Artificial Neural Network applied to classification of lightning discharge images

The development of video cameras created new ways of studying lightning discharges. The main goal of this project is to create a system using C Language and OpenCV libraries to detect images of lightning discharges stored in videos and classify them using an Artificial Neural Network (ANN). The developed system, can be split in two different modules: detection module and classification module. The detection module analyses recorded videos and, using OpenCV computer vision libraries and image processing techniques, detects if there are significant differences between frames in a sequence, indicating that something, still not classified, occurred. Whenever there is a significant difference between consecutive frames, two algorithms are used to analyze the frame image: brightness and shape algorithms. These algorithms detect shape and brightness of the event, removing irrelevant events like birds, as well as detecting the relevant events exact position, allowing the system to track it over time. The classification module calculates the event's number of discharges, and uses a neural network, which classifies them as horizontal or vertical lightning, and then save the event image. The Neural Network implemented using the backpropagation algorithm, and was trained with 42 training images , containing 57 lightning events (one image can have more than one lightning). An ANN with one hidden layer was tested with five different configurations, with layers having up to 50 neurons. This initial test showed that the best configuration, which achieved a success rate of 95%, with a layer with 20 neurons (33 test images with 42 events were used in this phaser). The developed system analyzed 20 video files, containing 63 lightning discharges previously manually detected. Results showed that all the lightning discharges were detected, many irrelevant events were unconsidered, and the event's number of discharges was correctly computed. The neural network classified the detected events with a success rate of 90%. The videos used in this experiment were acquired by seven video cameras installed in São Bernardo do Campo, Brazil, that continuously recorded lightning events during the summer. The cameras were disposed in a 360º loop, recording all data at a time resolution of 33ms. During this period, several convective storms were recorded.