Development of real-time video target recognition system based on convolutional neural network

Video image enhancement as the implementation of the basis of video target recognition, this paper in the linear grayscale transformation, histogram adjustment, histogram equalization of three video image enhancement technology based on comparative experiments to determine the linear grayscale transformation technology as the means of video image enhancement processing in this study. After the video image is effectively enhanced, the target recognition network structure is built by integrating the convolutional neural network, the target recognition algorithm is improved based on YOLO-V5, and the convolutional attention mechanism module CBAM is added to enhance the feature extraction ability of the video image, so as to construct a real-time video target recognition model, and develop a real-time video target recognition system. In order to understand the training state of the model before and after the improvement, model training experiments are carried out. With the growth of training time, the precision of this model is maintained at about 0.848 after 50 epochs, the recall rate reaches the first peak at the 80th epochs, and the average precision is close to the peak at the 45th epochs, which makes the model performance better. Applying the real-time video target recognition system in this paper to complete the task of real-time video target recognition of rabbits and young rabbits, the total time consumed by this paper’s system in the first, second, and third group of video image recognition is lower than that of the traditional system, and the recognition speed is faster than that of the traditional system by 0.09s, 0.059s, and 0.0377s, respectively.