SMALL TARGET DETECTION ALGORITHM BASED ON IMPROVED CONVOLUTIONAL NEURAL NETWORK

In order to solve the problems of small-scale target missing detection and imprecision in detection, an improved YOLO v4 algorithm model (F-YOLO v4) is proposed. The improved K-means clustering algorithm was used to cluster the data set, which made the size of anchor frame more suitable for target detection. The convolution within and between channels was separated by the combination of deep convolution and point by point convolution, so as to improve the original residual block. The channel attention mechanism was used to improve the backbone network, and the RFB module was added to the PANet network to increase the number of residual blocks. The feature extraction ability was enhanced, and the detection effect of small target was improved. The experimental results show that the average accuracy of F-YOLO v4 algorithm on KITTI data set reaches 93.67%, which is 1.52 percentage points higher than the original algorithm, and has higher accuracy than other mainstream networks.