JP7614994B2 - Water Quality Monitoring System - Google Patents

Description

Embodiments of the present invention relate to water quality monitoring technology.

Water treatment facilities use water quality meters such as turbidity meters to detect abnormalities (cloudiness) in treated water. There are two types of water quality meters: one that measures turbidity by placing the sensor part in the water being monitored, and one that measures turbidity by shining light on the surface of the water being monitored (measured liquid level) and measuring the light scattered from the water surface (liquid level).

JP 2017-207323 A

We provide a water quality monitoring system that uses images captured by a surveillance camera to achieve non-contact water quality monitoring and can accurately detect water quality abnormalities by suppressing the effects of changes in the state of the monitored object.

The water quality monitoring system of the embodiment includes a water quality abnormality detection unit that uses an object detection model to recognize a water surface area from a captured image of a water quality monitoring target output from a monitoring camera and determines water quality abnormality based on the state of the recognized water surface area, a monitoring result output unit that outputs the monitoring result , a learning data collection unit that stores learning data that is a set of measurement values of a water quality meter provided on the water quality monitoring target and monitoring images corresponding to the measurement values, a learning data selection unit that automatically sorts the monitoring images into two or more groups using a predetermined reference value corresponding to the measurement values, a training teacher data generation unit that displays the monitoring images by group on a display device, accepts a manual group sorting operation by an operator for the automatically sorted monitoring images, and generates teacher data based on the monitoring images by group sorted through automatic sorting and manual sorting, and a learning processing unit that performs a learning process of the object detection model by applying the generated teacher data . The predetermined reference value can be a reference value indicating an abnormality of the water quality monitoring target based on the water quality meter, or a reference value indicating a normality of the water quality monitoring target based on the water quality meter.

FIG. 2 is a functional block diagram of each device constituting the water quality monitoring system of the first embodiment. FIG. 2 is an explanatory diagram of an object detection model according to the first embodiment. FIG. 2 is an explanatory diagram of a water quality monitoring model according to the first embodiment. FIG. 2 is a diagram showing a processing flow of the water quality monitoring system of the first embodiment. FIG. 2 is a diagram showing a processing flow of the water quality monitoring system of the first embodiment, and is a diagram showing a processing flow taking into account the measurement value of a water quality meter. FIG. 2 is an explanatory diagram of collection of learning data and generation of teacher data by the learning device of the first embodiment. FIG. 2 is a diagram showing a processing flow relating to teacher data generation of the learning device of the first embodiment.

The following describes the embodiment with reference to the drawings.

First Embodiment
1 to 7 are diagrams for explaining the water quality monitoring system of the first embodiment.

One example of a water quality monitoring target to which the water quality monitoring system of this embodiment is applied is the wastewater treatment process of a water treatment facility. There are various types of wastewater treatment, but for example, in the case of wastewater treatment using the activated sludge method, raw water is introduced into a reaction tank (aeration tank) containing activated sludge, which is sludge containing microorganisms, and air is pumped in to stir the water. As a result, the microorganisms break down the dirt in the raw water, and fine dirt adheres to the microorganisms and becomes clumps that are easy to sink. The sludge (activated sludge) produced in the reaction tank is then settled in a settling tank and separated into treated water (supernatant) and sludge. The separated treated water is then filtered, and the remaining dirt is adsorbed by activated carbon in an activated carbon tower, and the water is discharged into a sewer or river after a final neutralization process (pH is brought to a neutral range).

A water quality meter is installed in the sedimentation tank to measure the SS concentration and monitor the water quality. The SS concentration refers to suspended solids (SS), or floating matter, which are substances suspended or dispersed in the water with particle sizes between 1 μm (0.001 mm) and 2 mm, and is expressed as the mass of suspended solids contained in 1 liter of water (mg/L).

However, water quality monitoring using a water quality meter also poses the following challenges. In the case of water quality meters that measure by placing the sensor part underwater in the sedimentation tank being monitored, the sensor part becomes susceptible to dirt (floating sludge, algae, etc.), so cleaning is required, and calibration work is also required to ensure the reliability of the measurement values.

In addition, in the case of water quality meters that measure by shining light on the water surface (measurement liquid surface) in a settling tank and measuring by capturing the scattered light from the water surface (liquid surface), since the light is reflected on the surface of the measurement liquid, it is affected by the color of the water and air bubbles, making it difficult to ensure measurement accuracy. In particular, from the perspective of measurement accuracy, the water surface needs to be stable, and measures such as providing a separate measurement tank have been taken to stabilize the water surface, but organic matter and scale components present in the wastewater adhere to the walls of the measurement tank and impede optical measurement, making maintenance of the measurement tank necessary.

Water quality monitoring using water quality meters in this way requires extremely difficult maintenance, and there are also issues with increased monitoring workload (for example, workers having to monitor by looking at security camera footage) in order to respond to inaccurate measurements that result from lack of maintenance.

The water quality monitoring system of this embodiment realizes non-contact water quality monitoring using images captured by a surveillance camera 1 that captures the settling tank, while suppressing the effects of changes in the state of the settling tank to enable accurate detection of water quality abnormalities.

Figure 1 is a functional block diagram of each device constituting the water quality monitoring system of this embodiment. As shown in Figure 1, the water quality monitoring system of this embodiment is composed of a monitoring device 100 and a learning device 200, and the monitoring device 100 uses a water quality monitoring model to detect water quality abnormalities in the water quality monitoring target captured by the monitoring camera 1. The learning device 200 collects and generates learning data (teacher data) and performs learning processes such as machine learning and deep learning of the trained model (water quality monitoring model) used by the monitoring device 100.

The monitoring device 100 and the learning device 200 are connected wirelessly or by wire to a monitoring camera 1 that captures the monitored object, and the monitoring image (monitoring video) output from the monitoring camera 1 can be input to each device 100, 200. In addition, the settling tank that is the monitored object is provided with a water quality meter 2 as in the conventional case, and the measurement value of the water quality meter 2 can be input to the monitoring device 100 and the learning device 200.

The surveillance camera 1 is, for example, a remotely controlled imaging device, and an operator can manually change the angle of view (zoom) and change the direction of the surveillance camera 1. An operator can remotely control the surveillance camera 1 as needed on a daily basis, change the direction of the camera, and adjust the angle of view using the zoom function, while checking the video and performing surveillance work. Note that even an imaging device with a fixed angle of view and camera direction can be applied to the water quality monitoring system of this embodiment.

<Monitoring device 100>
As shown in FIG. 1, the monitoring device 100 includes a monitoring control unit 110, a monitoring result output unit, and a storage unit 130, and the monitoring control unit 110 has a water quality abnormality detection unit 111.

The water quality abnormality detection unit 111 uses an object detection model to recognize water surface areas from images of the water quality monitoring target output from a surveillance camera, and functions as a water quality monitoring model that determines water quality abnormalities based on the state of the recognized water surface area.

The object detection model can be, for example, a known object detection AI model such as a Convolutional Neural Network (CNN), including R-CNN, Fast R-CNN, Faster R-CNN, and Mask R-CNN.

The object detection model of this embodiment is a trained model that has been trained using images that include the water surface area of a settling tank, and performs a check to see if there is an object in the image (object detection), and if so, whether the object is on the water surface (object recognition), outputting area information such as the coordinates of the recognized water surface. For example, when a surveillance image is input, features are extracted in the convolution layer, and the pooling layer makes it possible to handle the features with variable vertical and horizontal sizes, thereby grasping the "water surface area" in the image. The object detection model also outputs the features extracted when grasping the water surface area, and these are used to detect turbidity in the surveillance target, which will be described later.

Figure 2 is an explanatory diagram of the object detection model of this embodiment. As shown in Figure 2, the water surface area in the image becomes smaller as the water surface position of the settling tank becomes lower. In addition, as the water surface position becomes lower, the brightness of the water surface becomes darker due to the influence of the shadow cast by the tank wall. As the water surface becomes darker, it becomes difficult to distinguish the boundary between the tank wall and the water surface. On the other hand, as the water surface position becomes higher, the water surface area in the image becomes larger, it is less influenced by the shadow cast by the tank wall, and the water surface is relatively bright.

In this embodiment, a learning process is performed on the water surface area according to the water surface position and brightness in this way, and an object detection model is constructed. As a result, the water surface area (monitoring target) can be accurately grasped even if the water level fluctuates. From another perspective, for example, if the angle of view of the surveillance camera 1 is changed by remote control or the camera orientation is changed, a difference occurs in the physical appearance of the water surface area in the image, similar to the water level fluctuation, but the object detection model of this embodiment can accurately grasp the "water surface area" in the image.

Next, the water quality abnormality detection unit 112 performs turbidity detection based on the features of the water surface area output from the object detection model, and determines water quality abnormalities based on the state of the water surface area recognized by the object detection model.

Figure 3 is an explanatory diagram of turbidity detection in this embodiment. In the example of Figure 3, monitoring images 1 to 3 are illustrated showing three abnormal water quality states. In monitoring image 1, many impurities are floating over the entire water surface area, and the transparency of the areas where there are no impurities is low. In monitoring image 2, impurities are not floating on the water surface, but the transparency is low. In monitoring image 3, impurities are floating in part of the water surface area, and there are no impurities in the remaining part. Compared to monitoring image 0, which shows a normal water quality state, monitoring images 1 to 3 show different objects reflected in the water surface area and different shading of the water surface corresponding to the transparency due to differences in impurities and transparency.

In this embodiment, a reliability score is calculated for each monitoring image according to the SS concentration, i.e., a feature related to the degree of turbidity, and a water quality abnormality is determined for the recognized water surface area based on the reliability score (a score indicating the degree of certainty that the water quality is abnormal).

For example, surveillance image 1 has many impurities floating over the entire water surface area, and the transparency of areas where there are no impurities is low, so the feature of impurities and the feature of low transparency can be extracted, and a high reliability score is calculated (e.g., reliability score 0.96). On the other hand, surveillance image 2 can extract the feature of low transparency, but does not extract the feature of impurities, so the second highest reliability score after surveillance image 1 is calculated (e.g., reliability score 0.78). Surveillance image 3 has the feature of impurities floating in part of the water surface area extracted, and the remaining part has no impurities and is also highly transparent. For this reason, the reliability score is calculated to be lower than surveillance images 1 and 2 (reliability score 0.52).

In the above explanation, the confidence score is calculated as a value between 0 and 1 as an example, but it can be any value that indicates the degree of certainty that there is an abnormality, and is not limited to being calculated as a value between 0 and 1.

The water quality abnormality detection unit 112 performs a learning process to distinguish between impurity features and low transparency features based on features extracted from the object detection model, and calculates a reliability score according to the impurity features and/or low transparency features. If the calculated reliability score exceeds a specified value, that is, if the degree of turbidity exceeds a specified value, it determines that there is a water quality abnormality. Note that the specified value may be a threshold derived by a learning process based on teacher data, or a threshold preset by an operator, etc.

In this way, the water quality abnormality detection unit 112 of this embodiment is a water quality monitoring model that includes an object detection model, and uses the object detection model to identify the water surface area (mudky water surface) in the monitoring image, calculates a reliability score using the features of the water surface area extracted by the object detection model when identifying the water surface area, and uses the calculated reliability score to determine water quality abnormalities.

According to this embodiment, an object detection model is used to identify the water surface area of the monitored object from the surveillance image, making it possible to realize non-contact water quality monitoring using images captured by a surveillance camera, and to accurately detect water quality abnormalities even when affected by changes in the state of the monitored object (such as fluctuations in water level, changes in the surveillance camera's angle of view, or changes in the camera's orientation).

Figure 4 is a diagram showing the processing flow of the water quality monitoring system of this embodiment. The surveillance camera 1 outputs a surveillance image, and the surveillance device 100 accepts the output surveillance image (S101). The surveillance device 100 (surveillance control unit 110) performs water quality monitoring processing. Note that the interval at which the surveillance images are input and the timing at which the water quality monitoring processing is performed are arbitrary.

The monitoring device 100 performs an object detection process using an object detection model on the received monitoring image, and detects the water surface area in the monitoring image (S102). Furthermore, the monitoring device 100 performs a water quality abnormality detection (turbidity detection) process on the detected water surface area, and judges the water quality abnormality based on the reliability score (S103).

If the monitoring device 100 (monitoring result output unit 120) determines in step S103 that an abnormality exists (YES in S104), it outputs a monitoring result of "abnormal" including a reliability score (S105). On the other hand, if the monitoring device 100 determines in step S103 that an abnormality does not exist (NO in S104), it outputs a monitoring result of "no abnormality" including a reliability score (S106).

For example, in the example of FIG. 3, "NG" indicating "abnormality detected" and a reliability score of "0.96" are output as the monitoring result on the monitoring image. Also, if the monitoring result is "no abnormality", it can be configured to output, for example, "OK" indicating "no abnormality" and a reliability score as the monitoring result.

As a monitoring result output process, the monitoring device 100 (monitoring result output unit 120) can notify the monitoring results via the network by email to a preset destination, send a push notification to a specified device, superimpose the monitoring results including the reliability score on a display device that displays the monitoring image, or display the monitoring results on the display device as time-series data rather than on top of the monitoring image. The monitoring device 100 (monitoring result output unit 120) stores the monitoring results in the memory unit 130 (S107).

In the above explanation, a mode of outputting monitoring results including a reliability score was described, but the monitoring results may be configured to be output in a mode that does not include a reliability score. By looking at the reliability score, workers can grasp the degree of abnormality, which can lead to more accurate and rapid response to water quality abnormalities.

In particular, by outputting monitoring results that include a reliability score, it is possible to go beyond simply notifying whether water quality is normal or abnormal and capture "changes in water quality" over time, providing information that acts as a "water quality anomaly warning" that suggests the possibility of future water quality anomalies. This makes it possible to respond before water quality anomalies occur. Furthermore, while predicting future water quality anomalies by looking at monitoring images requires skilled and experienced personnel, the reliability score makes it possible to quantitatively grasp the state of water quality, providing appropriate support for workers' monitoring work.

Although a reliability score expressed as a numerical value is provided as an aspect of the monitoring result including the reliability score, this is not limiting. For example, the reliability score may be ranked in advance using a predetermined numerical range, and the monitoring result including the rank of the water quality abnormality may be output as another expression of the reliability score. In addition, the color of at least a part of the monitoring result may be changed depending on whether a water quality abnormality is determined or not, so that the water quality abnormality can be easily recognized, or the monitoring result may be output so that different effects are applied depending on whether the degree of abnormality is high or low when a water quality abnormality is determined.

Figure 5 shows the process flow of the water quality monitoring system of this embodiment, taking into account the measurement value (sensor value) of the water quality meter 2. Processes similar to those in Figure 4 are given the same reference numerals and will not be described.

In the example of FIG. 5, in addition to the monitoring image, the measurement value of the water quality meter 2 corresponding to the timing of capturing the monitoring image is input to the monitoring device 200 (S101a). Then, the monitoring device 200 (monitoring control unit 110) determines whether or not there is a water quality abnormality in the measurement value of the water quality meter 2, in other words, whether or not the measurement value of the water quality meter 2 exceeds a predetermined threshold value for water quality abnormality (S104a).

In step S104b, regardless of the result of the water quality abnormality determination by the water quality meter 2, a water quality abnormality determination process is performed for the water surface area extracted by the object detection process, similar to step S104 in FIG. 4. The monitoring device 200 then outputs the water quality abnormality determination result of step S104b as the monitoring result.

By configuring in this manner, even if an abnormality occurs in the measurement value of the water quality meter 2 due to lack of maintenance as described above, accurate water quality monitoring can be performed using the water quality monitoring model, and the occurrence can be grasped as an opportunity to perform maintenance on the water quality meter 2. In other words, if both the water quality monitoring model and the measurement value of the water quality meter 2 in this embodiment are judged to be normal, the water quality meter 2 is also operating normally, and if the water quality monitoring model is normal and the measurement value of the water quality meter 2 is abnormal (or if the water quality monitoring model is abnormal and the measurement value of the water quality meter 2 is normal), maintenance of the water quality meter (including failure) is required, and if both the water quality monitoring model and the measurement value of the water quality meter 2 are judged to be abnormal, it can be more accurately grasped that an abnormality has occurred in the water quality of the monitored object.

In the above description, the reliability score is an index showing abnormal water quality, but the invention is not limited to this and may be configured to calculate the reliability score as an index showing normal water quality and determine whether water quality is abnormal. Specifically, the water quality abnormality detection unit 112 performs a learning process to distinguish between impurity features and high transparency features based on the features extracted from the object detection model, and calculates a reliability score according to the impurity features and/or high transparency features. If the calculated reliability score is less than a specified value, that is, if the transparency level is lower than the specified value (the turbidity level is higher than the specified value), it can determine that the water quality is abnormal.

The water quality abnormality detection unit 111 can therefore calculate an index indicating that the water quality is normal as a reliability score based on the features of the turbidity state of the water surface area extracted by the object detection model, and determine whether the water quality of the recognized water surface area is abnormal based on the reliability score.

The water quality abnormality detection unit 112 can also be configured to calculate a first reliability score of an index indicating water quality abnormality and a second reliability score of an index indicating normal water quality, and perform water quality abnormality determination using both reliability scores. For example, when the first reliability score exceeds a first specified value and the second reliability score is lower than a second specified value, it can be determined that the water quality of the water surface area is abnormal. In this case, the water quality abnormality detection unit 112 can be configured to include a first trained model generated through a learning process based on "abnormality" from the perspective of water quality abnormality, and a second trained model generated through a learning process based on "normality", and a water quality monitoring model can be constructed by connecting these first trained model and second trained model in parallel or in series.

<Learning device 200>
As shown in FIG. 1, the learning device 200 includes a learning data collection unit 210, a learning data selection unit 220, a learning teacher data generation unit 230, a learning processing unit 240, and a storage unit 250.

Figure 6 is an explanatory diagram of the collection of learning data and the generation of teacher data by the learning device 200. Figure 7 is a diagram showing the process flow for generating teacher data by the learning device 200.

As shown in FIG. 6, the learning data collection unit 210 stores learning data in the memory unit 250, which is a set of measurement values (sensor values) of the water quality meter 2 installed on the water quality monitoring target and monitoring images corresponding to the measurement values. At this time, if the surveillance camera 1 and the water quality meter 2 are connected to the learning device 200 via a network, the learning data collection unit 210 can accept the learning data by data transmission via the network. Note that the surveillance camera 1 and the water quality meter 2 do not necessarily need to be connected via a network, and the learning device 200 only needs to be configured to be able to accept the monitoring images and measurement values output from them as learning data.

The learning data selection unit 220 performs a sorting process that automatically sorts the collected monitoring images into two or more groups using a predetermined reference value corresponding to the measurement value of the water quality meter 2. Specifically, a reference value indicating an abnormality in the water quality monitoring target based on the water quality meter 2 or a reference value indicating a normality in the water quality monitoring target based on the water quality meter is set as a predetermined reference value in advance and stored in the storage device 250 (various data 252). The learning data selection unit 220 then compares the measurement value included in the learning data with the set reference value, and performs a temporary automatic sorting (provisional sorting) into a group of monitoring images that exceed the reference value and a group of monitoring images that are below the reference value.

The learning teacher data generation unit 230 displays the surveillance images by group on a display device and accepts manual group sorting operations by an operator for the automatically sorted surveillance images. In other words, the learning teacher data generation unit 230 allows an operator to visually check the surveillance images included in each provisionally sorted group, and provides a manual sorting function by an operator to remove surveillance images from a group or switch groups as necessary. The learning teacher data generation unit 230 generates (accumulates) the surveillance images by group sorted through automatic sorting (provisional sorting) and manual sorting as teacher data.

Note that, although the learning device 200 of this embodiment is described based on "abnormality" from the perspective of water quality abnormalities, it may also be configured to sort learning data and generate teacher data based on "normality."

The learning processing unit 240 applies the generated training data to perform a training process for a water quality monitoring model including an object detection model. The training process for the object detection model can be configured to perform a segmentation process in advance on the training data (monitoring image), for example. Publicly known methods can be appropriately adopted for these training processes and training data.

In this way, the learning device 200 of this embodiment has a function of provisionally sorting collected learning data based on the measurement values of the water quality meter 2.

The learning process requires a huge amount of learning data, and it is also necessary to select the teacher data (normal/abnormal). Conventionally, workers would check each image data and select it, but this was an inefficient way of working. In particular, when the frequency of water quality abnormalities is low, it takes a lot of time to select images of water quality abnormalities from the huge number of images.

Therefore, in this embodiment, the measurement values of the water quality meter 2 are collected together with the monitoring images, and a predetermined reference value corresponding to the measurement value of the water quality meter 2 is set (S201). Then, a primary diagnosis is performed to automatically separate monitoring images that exceed the threshold into an abnormal image group and monitoring images that do not exceed the threshold into a normal image group (S202).

The provisionally sorted monitoring images are displayed on a display device by group (S203), and finally, an operator manually checks each monitoring image and selects it as training data (S204). The operator only needs to check each image in the abnormal image group with the understanding that it is an image of abnormal water quality, and check each image in the normal image group with the understanding that it does not contain any images of abnormal water quality. In other words, work efficiency is improved and the time spent on image selection can be significantly reduced compared to selecting images of abnormal/normal water quality from a large number of images that contain a mixture of abnormal and normal water quality. The monitoring images sorted by group through automatic and manual sorting are then generated (stored) as training data (S205).

Although the embodiment has been described above, the monitoring device 100 and the learning device 200 may be configured as a single device, and configured as a monitoring device equipped with a learning system. The devices 100 and 200 can also be constructed as a cloud-based service provision form. In other words, the monitoring images output from the monitoring camera 1 and the measurement values output from the water quality meter 2 are configured to be transmitted to the monitoring device 100 and the learning device 200 via an IP network. This allows the learning data to be collected on the cloud side, and the monitoring results can be output (provided) to monitoring equipment (monitoring terminals, etc.) that can be viewed by workers. The learning device 200 may also be configured in a form in which the function of collecting and storing learning data and the learning function are configured as separate devices.

The devices 100 and 200 are computer devices equipped with the computational functions, storage functions, and communication functions of a server device or the like. The hardware configuration may include a memory (main storage device), operation input means such as a mouse, keyboard, touch panel, and scanner, output means such as a printer, and an auxiliary storage device (hard disk, etc.).

Furthermore, each function of the present invention can be realized by a program, and a computer program prepared in advance to realize each function is stored in an auxiliary storage device, and a control unit such as a CPU reads the program stored in the auxiliary storage device into a main storage device, and the control unit executes the program read into the main storage device, thereby causing the computer to operate the functions of each unit of the present invention. On the other hand, each function of devices 100 and 200 can be configured as a separate device, or a computer system can be configured by connecting multiple devices directly or via a network.

The above program can also be provided to a computer in a state in which it is recorded on a computer-readable recording medium. Examples of computer-readable recording media include optical disks such as CD-ROMs, phase-change optical disks such as DVD-ROMs, magneto-optical disks such as MO (Magnet Optical) and MD (Mini Disk), magnetic disks such as floppy (registered trademark) disks and removable hard disks, and memory cards such as Compact Flash (registered trademark), Smart Media, SD memory cards, and memory sticks. Also included as recording media are hardware devices such as integrated circuits (IC chips, etc.) that are specially designed and configured for the purposes of the present invention.

Although an embodiment of the present invention has been described, this embodiment is presented as an example and is not intended to limit the scope of the invention. This new embodiment can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are included in the scope of the invention and its equivalents described in the claims.

REFERENCE SIGNS LIST 1 Monitoring device 2 Water quality meter 100 Monitoring device 110 Monitoring control unit 111 Water quality abnormality detection unit 120 Monitoring result output unit 130 Memory unit 200 Learning device 210 Learning data collection unit 220 Learning data selection unit 230 Learning teacher data generation unit 240 Learning processing unit 250 Memory unit 251 Learned model 252 Various data

Claims (4)

a water quality abnormality detection unit that uses an object detection model to recognize a water surface area from a photographed image of a water quality monitoring target output from a monitoring camera and determines water quality abnormalities based on the state of the recognized water surface area;
a monitoring result output unit that outputs the monitoring result;
a learning data collection unit that stores learning data that is a set of measurement values of a water quality meter provided in a water quality monitoring target and a monitoring image corresponding to the measurement values;
a learning data selection unit that automatically classifies the surveillance images into two or more groups using a predetermined reference value corresponding to the measurement value;
a training teacher data generating unit that displays the grouped surveillance images on a display device, accepts manual group sorting operations by an operator on the automatically sorted surveillance images, and generates teacher data based on the grouped surveillance images sorted through the automatic and manual sorting;
A learning processing unit that performs a learning process of the object detection model by applying the generated teacher data,
A water quality monitoring system characterized in that the predetermined reference value is a reference value that indicates an abnormality in the water quality monitoring target based on a water quality meter, or a reference value that indicates a normality in the water quality monitoring target based on a water quality meter . The water quality abnormality detection unit calculates a reliability score based on the feature amount of the turbidity state of the water surface area extracted by the object detection model, and determines the water quality abnormality of the recognized water surface area based on the reliability score;
The water quality monitoring system according to claim 1 , wherein the monitoring result output unit outputs the monitoring result including the reliability score. The water quality monitoring system according to claim 1 or 2, characterized in that the angle of view or the direction of the surveillance camera can be changed by manual operation by an operator. A program executed by a computer,
A first function of using an object detection model to recognize a water surface area from a photographed image of a water quality monitoring target output from a surveillance camera, and determining water quality abnormalities based on the state of the recognized water surface area;
A second function of outputting the monitoring result;
a third function of storing learning data that is a set of measurement values of a water quality meter provided in the water quality monitoring target and monitoring images corresponding to the measurement values;
a fourth function of automatically classifying the surveillance images into two or more groups using a predetermined reference value corresponding to the measurement value;
a fifth function of displaying the grouped surveillance images on a display device, accepting manual group sorting operations by an operator for the automatically sorted surveillance images, and generating teacher data based on the grouped surveillance images sorted through the automatic sorting and manual sorting;
A sixth function of performing a learning process of the object detection model by applying the generated teacher data to the computer ;
The program, characterized in that the predetermined reference value is a reference value indicating an abnormality in the water quality monitoring target based on a water quality meter, or a reference value indicating a normality in the water quality monitoring target based on a water quality meter .

Images