Research on object detection method based on neural network

In view of the existing target detection algorithms are trained and evaluated on the general data set, while the targets in t he general data set account for a large proportion of the images, these target detection algorithms are directly used for the poor detection effect of weak targets, and the weak target detection method based on PVANet is proposed. The ability of this method to identify small targets meets the requirement of the intelligent perception system to identify weak targets. Combined with pruning, quantification, weight value clustering, Huffman coding and other methods, deep neural network lightweight design, compress the network size, reduce the number of parameters, and achieve higher computational performance under limited computing resources. This method has great advantages in capturing small and large targets in the image, while improving the method of generating target candidate boxes, which can effectively detect weak targets in the image.