A Neural Network-Based Intelligent Image Target Identification Method And Its Performance Analysis

The image sensor-based target identification is one of the important for vehicle identification and intelligent transportation applications. Based on neural network classification algorithms, one vehicle identification method is proposed. The main work includes: (1) designing the parameter extraction of vehicle shape based on Sobel operator and mathematical morphology; (2) introducing a vehicle feature extraction method based on "I"-shape model; (3) designing a momentum backpropagation algorithm with the smallest error and a variable learning rate of an exponential function, verify the algorithm by simulation experiments. The proposed algorithm combines the advantages of the algorithm with additional momentum and variable learning rate while the convergence speed is accelerated and jumps out of local minimum points. The results of simulation experiments show that the proposed algorithm can quickly and effectively identify the various types of vehicle models.