AU2008200979B2 - Lidar system controlled by computer for smoke identification applied, in particular, to early stage forest fire detection - Google Patents

Description

DESCRIPTION LIDAR system controlled by computer for smoke identification applied, in particular, to early stage forest fire detection FIELD OF THE INVENTION The present invention -relates to a LIDAR (Light Detection and Ranging) system controlled by computer for smoke identification applied, in particular, to early stage forest fire detection which includes a scanning laser beam device and a data processing unit monitored by a neural network. More in particular, the present invention relates to a method of fire detection via analysis of a part of the scanning laser beam radiation backscattered by smoke by identification of eventual smoke plumes signatures with software based on a neural network algorithm; following up the operation of this software and activating alarm if smoke is detected; controlling the scanning motors in order to realise the programmed scanning procedure; modifying the scanning procedure and other system parameters, such as the laser pulse energy and the number of accumulated LIDAR returns as required by the operator in dependence of the atmospheric visibility at current weather conditions.

2 BACKGROUND OF THE INVENTION Forest fire detection may be done using passive or active methods. Active methods use LIDAR techniques while passive methods use either image processing, which allows the detection of smoke and flames, or static detectors, which measure flame radiation intensity, temperature, smoke density and composition. Active systems have several advantages, mainly the detection at larger distances and during the early stage of the fire. While most of the available systems are passive, this invention is concerned with a method and an active system for forest fire detection using LIDAR. Only a few former inventions use active systems for forest fire detection. However, they use uniaxial LIDAR configurations, which make them inefficient, because the emitted beam and the backscattered radiation travel along the same path. The emitted and received radiation must be separated with by a polariser, decreasing the optical efficiency. Furthermore, the system is more complicated, the scanning process is slower and the signal processing has low efficiency. Most patents, like DE4026676, US5734335, US5422484, US5218345, US5049756, US5751209, and US5168262, are related to passive systems that use static detectors. On the other hand, patents EP0984413, EP818766A1, US5557260, and W09408660A1 describe special vision systems adapted to fire detection. Only two patents, US4893026 and EP978718A1, are devoted to active detection.

3 The USA patent US4893026 describes a LIDAR system capable of locating an object likely to diffuse back part of an incident laser beam. However, this LIDAR system is uniaxial, that is, the laser beam and the backscattered radiation pass along the same trajectory and it is necessary to use polarised radiation to separate the emitted and received light. The system has an optical unit that expands the laser beam, collects the backscattered radiation and separates it from the emitted beam and directs it to the detector for measuring. The system uses a pulsed Nd:YAG laser with an energy of 300 mJ per pulse and a pulse repetition rate between 5 Hz and 30 Hz. Angular scanning is performed by uniform rotation of the output mirror without any optimisation concerning the surrounding landscape. The use of a uniaxial LIDAR limits the detection efficiency due to polarisation losses and makes the system very expensive and difficult to maintain and operate. On the other hand, the low flexibility of the scanning 4 system limits its use in hilly landscapes. Without any experimental evidence, the authors claim a detection range of 20 km. On the contrary, the LIDAR proposed in the present patent has a biaxial optical system and a scanning procedure optimised according to the features of the surrounding landscape. Without polarisation of radiation, the proposed system is more efficient and easier to build and operate, and also enables fires to be detected at an earlier stage and at larger distances. Rui Vilar and Alexander Lavrov ("Estimation of required parameters for detection of small smoke plumes by LIDAR at 1.54 pm", Appl. Phys. B vol. 71, no. 2, p. 225-229, 2000) described the LIDAR for forest fire detection. They suggested using an eye-safe lidar for detection of small forest fire. Theoretical range of detection is estimated to be 10 km. No embodiment of the ldar system or automatic detection method is described. A method of forest fire detection is discussed by Begofia C. Arrue, Anibal Ollero, J. Ramiro Martinez de Dios ("An Intelligent System for False Alarm Reduction in Infrared Forest Fire Detection," IEEE Intelligent Systems, vol. 15, no. 3, p. 64-73, 2000). In this reference smoke plumes are automatically identified with the False Alarm Reduction system, which make use of image processing techniques and artificial neural networks. In order to design a fuzzy expert rule base for decision making, the system harnesses additional information from meteorological sensors, geographical information databases, etc. and taking advantage of the information redundancy from visual and infrared cameras through a special matching process. Although using neural networks, this detection method differs from the methods presented here as passive imaging, rather than the active lidar technology, is implemented for detection. The European patent EP978718Al is related to a biaxial LIDAR for smoke detection that does not have any scanning mechanism. Being intended for use in closed spaces, like tunnels and underground parking surveillance, the system has mirrors to reflect or scatter the laser beam in order to cover the 5 surveillance area in a proper way. The mirrors may be flat or curved and they are usually fixed. The processing algorithm is inadequate for covering large areas. In contrast to the present invention, the system does not possess automatic scanning system and an automatic smoke recognition system. The previously mentioned passive systems of static detectors operate on the basis of some optimised distribution of the sensors across the area under surveillance. The sensors detect particles, smoke, or thermal radiation due to burning. The passive systems based on video-camera supervision acquire images of the surveillance area that are compared with the reference images. Image processing uses several methods, such as filtering, overlapping, tracing, colour comparison, cluster recognition, etc. Some of the detecting systems have real time signal processing and use statistical or neural-network algorithms to improve the detection. European patent EP0375640 discloses a method for determining the characteristics (dimensions, refractive index, etc.) of diffusion sources interacting with a wave. In particular, it discloses a biaxial LIDAR station comprising of 4 mJ, 250 Hz laser source generating simultaneously at two wavelengths, 598.8 and 299.4 nm. The receiver system consisted of a 30 cm diameter telescope, an iris, various filters for cutting out ambient light and a photomultiplier. The random characteristic of a distributed scattering object (in particular, a smoke plume) is defined on the basis of a model describing the scattering of a plane wave by a stochastic set of wave-scattering objects. The subject matter of the present invention differs from the disclosure of EP0375640 in the identification method (a neural network rather than a stochastic scattering model) and in that a computer controlled beam scanning is performed.

C WRPotblDCC\AX\2850422 1.DOC.9M4/2lO0 6 US5225810 discloses a fire detection device for discrimination smoke and flame based on changing characteristics of optically measured distances. This fire detection system is very similar to rangefinder, and needs for its reliable work so-called "referenced object", for example, a wall situated behind the fire. In the case when smoke is absent the rangefinder measures the distance to the wall on the basis of the lapse of time from the emission to the detection of the radiation pulse. No neural network system for smoke recognition is disclosed; additionally, as the detection method requires a permanent presence of a static reference object within the LIDAR field of view, no scanning system is implemented. Embodiments of the present invention propose a biaxial active system that uses backscattered laser radiation. The sensitivity of this method is higher than that of the passive methods allowing fires to be located at early stages, when the passive detection is not yet possible. On the other hand, the system is cheaper and, from the viewpoint of construction and operation, simpler than the uniaxial LIDAR systems. As opposed to the lidar described by Vilar and Lavrov and those disclosed in patents EP978718A1, US-A-5 343284, EP0375640 and US5225810, the proposed system enables the smoke plumes to be automatically identified by means of a neural network. SUMMARY OF THE INVENTION According to the present invention, there is provided a method for detecting and locating smoke based on a LIDAR detector, which comprises: - emitting a pulsed beam (1) with an energy of 1 pJ - 1 J per pulse, repetition rate of 0.5 Hz - 20 kHz, and C:MRPonbl\DCC\AXL\2850422 _,DOC-9104/20 10 7 wavelength in the range 0.2-12 pm, produced by a laser source (2); - collecting the radiation backscattered (3) by smoke via a telescope (4) that does not share the optical axis of the probing beam (biaxial architecture); - measuring the radiation (3) collected by the telescope (4) with a photo detector (5) provided with an optical filter (6) centred at the wavelength of the laser radiation and with 0.1 - 10 nm spectral width for suppression of background radiation and improving signal to noise ratio, and a diaphragm (7) to control the detector aperture; - scanning the laser beam (1) and radiation receiver set-up (8) under computer (9) control according to the given topography; - transforming the detector signal using an analog-to digital converter (10) and sending the resulting data to a memory unit (11); - accumulating the LIDAR return signals in the memory unit (11); characterised by - processing and analysing the LIDAR signals and identifying smoke plumes signatures and locating the smoke plume with software based on a neural network algorithm (12), in which the neural network is simulated or implemented as a co-processor; - controlling the detector with high-level software (13) by: (i) monitoring the output of the neural network software (12) and activating an alarm (14) if a smoke plume signature is detected; C :\RPonbl\DCC\AXLU\Z5422 I DOC-9A)4/201 0 8 (ii) controlling a step motor detector scanning unit (15) in order to perform a pre-programmed scanning procedure, optimised with respect to the azimuth and elevation angles to ensure maximum area coverage; (iii) if necessary, modifying the scanning procedure, laser pulse energy and number of accumulated LIDAR returns according to the ground topography and current atmospheric and weather conditions. Embodiments of the system of the present invention provide eye-safe conditions by using an eye-safe laser wavelength, laser beam, or by decreasing the laser power while simultaneously increasing the pulse repetition rate. An optical parametric oscillator, an optical crystal or a system of optical crystals, or a Raman gas cell can be used to change the laser wavelength and provide eye-safety, if required. In an alternate embodiment of the present invention, two or more LIDAR detectors can be used to monitor adjoining and/or partially overlapping surveillance areas, functioning in a network with central data-processing, control, and decision-making unit and connected to common reference databases. In this embodiment, if one LIDAR detector detects an object likely to be a smoke plume, it sends an alarm signal to the central unit that directs the scanning of the neighbouring LIDAR detectors to the suspicious area to confirm the event and, if the presence of smoke is confirmed, generates the final alarm. Optionally, several laser wavelengths can be used simultaneously. In this manner it is possible to analyse the chemical properties of the detected plume by differential absorption LIDAR (DIAL) with a biaxial architecture.

C:\NRPonbl\DCCXL\250422_ 1DOC-9/04/2010 9 Embodiments of the method of the present invention are of potential application, in particular, to early forest fire detection and localization where the data-processing software provides relevant characteristics of the fire. The proposed system is simple and does not require polarised radiation, so its efficiency is much higher, mainly due to no overlapping between the emitted and backscattered radiation paths. One computerised unit using neural networks 10 processes the backscattered radiation signal due to particles that cross the laser beam and identifies the smoke plume signatures resulting from the fires using neural networks. To survey wider areas under surveillance it is possible to connect use and synchronise two or more synchronized LIDAR stations. If the stations are running in a network it is possible to evaluate and confirm events by triangulation, increasing the system efficiency and decreasing the risk of false alarms. BRIEF DESCRITPION OF FIGURES The present invention will be described with reference to the attached drawings where: Figure 1 shows the operation of a surveillance network composed by two LIDAR stations and a command centre according to the method of the present invention, also as a block diagram. Figure 2 shows typical LIDAR signals obtained in experimental conditions (a-f) of Table 2. Parameter n indicates the number of LIDAR returns, accumulated in the signal curve. DETAILED DESCRIPTION OF THE INVENTION The present invention describes a method and a system for detection and localisation of smoke, particularly suitable for early detection of forest fires. This active detection method is based on LIDAR technology. It differs from the method proposed in the USA patent US4893026 (1985) in at least three aspects: 1) The proposed LIDAR station has a biaxial optical architecture, in which the emitted and the backscattered radiation travel along different optical paths. For this reason the laser beam may be unpolarised and losses due to separation of emitted and collected radiation via polarisation are avoided. This difference makes the system simpler and more efficient than the uniaxial architecture of the above-mentioned patent.

11 2) The computer control of the LIDAR station allows optimising the scanning procedure for the given topography and other characteristics of the area under surveillance. For example, a high sampling density may be used in forest or inhabited regions. On the other hand, large rocky areas and lakes may be covered with a much lower sampling density. The consideration of statistical data on the previous fire occurrences allows optimising the scanning procedure even more. 3) The use of a neural network diminishes the probability of false alarms. For example, signals due to nearby chimney smoke can be included in the "no-alarm" part of the set with which the neural network is trained. Due to the nature of neural-network algorithms, no sophisticated preliminary signal pre-processing (such as smoothing, range adjustment, and logarithmic representation) is needed, which considerably simplifies signal processing and increases the speed of response. In the simplest configuration as shown in Fig. 1, the LIDAR station proposed in the invention comprises a laser 2, an optical system with a receiver 8 to capture the backscattered radiation 3 and a detector 5, an analog-to-digital-converter 10, and a computer 9 for signal processing and analysis 11, 12 that operates the whole LIDAR station and performs, by means of specific software 13, external communications 14. The optical receiver 8 includes a telescope 4, an adequate filter 6 for the laser radiation used, and one or more diaphragms 7 while the detector 5 may be a photomultiplier, photoconductive element or avalanche photodiode. The laser 2 periodically emits radiation pulses 1 with a fixed wavelength in the range 0.2-12 pm. The radiation 1 wavelength should be selected within one of the spectral windows of high atmospheric transmittance. The transmission window of the receiver's filter 6 should be centred in the laser 12 wavelength with a bandwidth from 0.1 to 10 nm. This optical element 6 is used for filtering the background radiation. The pulse 1 energy may be selected between 1 pJ and 1 J, with a repetition rate in the range 0.5 Hz - 20 kHz. If necessary, the laser radiation wavelength may be changed with the help of an optical parametric oscillator, a nonlinear crystal, or a Raman cell. If the energy per square meter exceeds 5x10 3 J/m 2 and the wavelength is in the band of high sensibility of the human eye (0.4-1.4 pm), a beam expander must be used. The data acquisition and signal pre-processing unit 9 comprises an analog-to-digital converter 10 (ADC) and other hardware necessary for signal treatment. In order to increase the signal to noise ratio, signals from successive laser pulses are accumulated. The specific number of signals 3 to accumulate depends on the laser pulse 1 energy, the repetition rate, and the wavelength. After accumulation, the resulting signal is analysed by the neural network 12. The desired LIDAR signal recognition algorithm is derived by the neural network 12 itself from examples, which form the training set. For the "no-alarm" situation the training set includes a real scene of the surveillance area with all its peculiarities (hills, vegetation, smoking chimneys, etc.) for different weather conditions. For the "start-alarm" condition, the training set can combine computer-generated scenes with real fire signals. When the LIDAR station detects smoke, the processing unit reports the characteristics of the fire, namely, the distance and the angular coordinates, and triggers the alarm 14. If a differential absorption LIDAR (DIAL) (based on the same biaxial architecture) is used, it is also possible to provide information on the chemical composition of the smoke. When the surveillance system comprises only one LIDAR station, it 13 continuously scans the landscape by changing, via computer controlled step motors 15, the azimuth and elevation angles. If a signal likely to be a smoke signature is detected, the system can re-scan the suspicious area in a more accurate way, in order to confirm the presence of the smoke plume. In order to cover a large surveillance area, several LIDAR stations may be integrated in a network. In this situation, if one station detects the a smoke plume, this event is reported to the automatic surveillance centre that instructs the neighbouring LIDAR stations to interrupt temporarily their routine scanning procedures and to verify the suspicious area for alarm validation. Finally, the surveillance centre analyses signals from different stations and, if other stations confirm the existence of smoke, emits the alarm signal. Although the signal processing algorithms are very important for alarm detection, the signal to noise ratio (SNR) is a fundamental criterion for preliminary evaluation of the detection quality. Several experiments made by the authors have demonstrated that it is possible to detect small smoke plumes produced by campfires with a burning rate as little as 0.02 kg/s at distances as large as 6.5 km (notably, these plumes cannot be observed from the LIDAR position even with standard binoculars). The SNR of the detected plume signatures were in the range 50 160, depending on the atmospheric conditions and the background radiation. EXPERIMENTAL RESULTS In order to test the viability of the present invention, a LIDAR station with the characteristics presented in Table 1 was designed, built, and tested.

14 Table 1 - Parameters of the LIDAR station used in the experiments Units of Parameter measurement Value Laser: flashlamp-pumped, water-cooled, Q-switched Nd:YAG repetition rate Hz 12 pulse duration ns 10 beam divergence mrad < 0.5 operating wavelengths pm 0.532 1.064 pulse energy mJ 30 90 Total transmitter efficiency % 90 90 Receiver: Cassegrainian telescope, lens diameter 30 cm, focal length 156.2 cm effective area T 2 0.0678 full angle of field of view mrad 0.9 Efficiency % 64 70 Bandwidth nm 4.8 5 photomultiplier FEU-83 with Peltier cooling dark current A 4.10~ Gain -105 photocathode responsivity mA-W' 0.7 0.3 Data acquisition system IBM-compatible PC with ADC ISA board Range km 1 - 30 sampling distance (6.25 MHz) m 24 on-board data buffer Kbyte 64 The experiments were made in Alentejo, in the south of Portugal, between the 27th of September and the 7th of October 2000; the experimental conditions are described in Table 2.

15 Table 2 - Set of experimental conditions Diameter Burning ID Date Fuel of the rate fire site Wood m kg/s a 04.10.00, olive 0.8 0.018 night b 05.10.00, Cedar tree 1.0 0.028 day c 05.10.00, Olive 0,8 0.025 night d 06.10.00, Cedar tree 1.2 0.025 day e 06.10.00, Olive 0.9 0.023 night f 07.10.00, Cedar tree 1.2 0.033 day During this period, more than 400 LIDAR signals were obtained. Some of them are illustrated in Figure 2. In each experiment about 100 kg of wood were burnt during 50-90 min. Some experiments were made in daylight conditions (b, d and f), while others were performed during the night (a, c and e) in less favourable conditions characterised by higher humidity and in the presence of haze. The values of SNR for different experimental conditions are presented in Table 3. Table 3 - Signal-to-noise ratio measured in different experimental conditions Number of Distanoe Signal-to Experimental accumulated Wavelength to the noise ratio conditions LIDAR plume (SNR) signals AM km a 128 1.064 3.90 90 a 128 0.532 3.90 70 b 128 0.532 3.90 130 C 4 0.532 4.66 23 c 8 0.532 4.66 32 a 16 1.064 4.66 8.2 o 256 0.532 4.66 89 a 128 0.532 6.45 49 .f 256 0.532 6.45 160 C:WRPOnbl\C\X\25[4221 LDOC-9 /2010 16 In all the situations the signal is so strong that no further signal processing is needed to provide evidence of the smoke-plume presence. The field tests clearly demonstrated that it is possible to detect smoke plumes of forest fires in early stages, even when the burning rate is as low as 0.02 kg/s, up to a distance greater than 6.5 km. Calculations performed by the authors on the basis of LIDAR modelling and experimental tests demonstrated that in the case of satisfactory atmospheric conditions and with a signal-to-noise ratio greater than 5 the detection range of the system exceeds 20 km. Throughout this specification and the claims which follow, unless the context requires otherwise, the word "comprise", and variations such as "comprises" and "comprising", will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps. The reference in this specification to any prior publication (or information derived from it), or to any matter which is known, is not, and should not be taken as an acknowledgment or admission or any form of suggestion that that prior publication (or information derived from it) or known matter forms part of the common general knowledge in the field of endeavour to which this specification relates. The reference numerals in the following claims do not in any way limit the scope of the respective claims.

Claims (7)

1. A method for detecting and locating smoke based on a LIDAR detector, which comprises: - emitting a pulsed beam (1) with an energy of 1 pJ - 1 J per pulse, repetition rate of 0.5 Hz - 20 kHz, and wavelength in the range 0.2-12 pm, produced by a laser source (2); - collecting the radiation backscattered (3) by smoke via a telescope (4) that does not share the optical axis of the probing beam (biaxial architecture); - measuring the radiation (3) collected by the telescope (4) with a photo detector (5) provided with an optical filter (6) centred at the wavelength of the laser radiation and with 0.1 - 10 nm spectral width for suppression of background radiation and improving signal to noise ratio, and a diaphragm (7) to control the detector aperture; - scanning the laser beam (1) and radiation receiver set-up (8) under computer (9) control according to the given topography; - transforming the detector signal using an analog-to digital converter (10) and sending the resulting data to a memory unit (11); - accumulating the LIDAR return signals in the memory unit (11); characterised by - processing and analysing the LIDAR signals and identifying smoke plumes signatures and locating the smoke plume with software based on a neural network algorithm (12), in which the neural network is simulated or implemented as a co processor; 18 - controlling the detector with high-level software (13) by: (i) monitoring the output of the neural network software (12) and activating an alarm (14) if a smoke plume signature is detected; (ii) controlling a step motor detector scanning unit (15) in order to perform a pre-programmed scanning procedure, optimised with respect to the azimuth and elevation angles to ensure maximum area coverage; (iii) if necessary, modifying the scanning procedure, laser pulse energy and number of accumulated LIDAR returns according to the ground topography and current atmospheric and weather conditions. 19

2. Method according to claim 1, characterised by eye-safe conditions provided by an eye-safe laser wavelength, the laser beam, or decreasing the laser power while simultaneously increasing the pulse repetition rate.

3. Method according to claims 1 and 2, characterised by the use of an optical parametric oscillator, an optical crystal or a system of optical crystals, or a Raman gas cell to change the laser wavelength and provide eye-safety.

4. Method according to the preceding claims characterised by using two or more LIDAR detectors to monitor adjoining and/or partially overlapping surveillance areas, functioning in a network with central data-processing, control, and decision making unit and connected to common reference databases: wherein if one LIDAR detector detects an object likely to be a smoke plume it sends an alarm signal to the central unit that directs the scanning of the neighbouring LIDAR detectors to the suspicious area to confirm the event, and, if the presence of smoke is confirmed, generates the final alarm.

5. Method according to the preceding claims characterised by using simultaneously several laser wavelengths in order to analyse the chemical properties of the detected plume by differential absorption LIDAR (DIAL) with a biaxial architecture.

6. Method according to the preceding claims characterised by its potential application, in particular, to early forest fire detection and localisation where the data-processing software provides relevant characteristics of the fire.

7. A method substantially as hereinbefore described with reference to the accompanying drawings.