Messa a punto di un sistema Lidar/Dial per l'allerta precoce degli incendi e per la minimizzazione dei falsi allarmi mediante lo sviluppo di una Rete Neurale

The use of a Lidar/Dial based system shows considerable advantages in the early forest fire detection, compared to the widely diffused infrared-cameras based passive detection methods, especially in order to minimize false alarms. We developed a compact mobile Lidar system based on both a Nd:YAG Q-switched laser source operating at the three wavelengths of 1064, 532 and 355 nm and on a CO2 laser source operating over about 60 laser lines in spectral range between 9 and 11 nm. The system is fully equipped for the early forest fire detection. A first evaluation of the atmospheric water vapour concentration increment (compared to the local standard value) has been estimated by numerical simulations. The results have been compared to experimental measurements carried out within a dedicated cell by the Lidar/Dial system, burning up several vegetable fuels. The above-mentioned Lidar/Dial system has been set-up and optimized by in-cell experimental measurements of the known smoke backscattering coefficients. Experiences were not limited to the in-cell measurement: field observations have been carried out, starting controlled fires, to demonstrate the system efficacy. Once the system laser source and data acquisition have been tested and optimized, we worked on the signal processing unit, developing a neural network based algorithm to implement an automatic smoke-plume recognition code. Several networks and learning rules have been compared and tested to obtain both the fire presence and absence conditions analyzing the Lidar returned signal. The alarm condition is returned with a fire distance information, calculated in the pre-processing stage of raw lidar signal. The pre-processing algorithm selects suspicious backscattering peaks and makes them unbiased and scale-independent. The resulting patterns can be successfully classified as corresponding to alarm or no-alarm conditions.