Lightweight Improved CenterNet Algorithm for Small Target Detection
In order to improve the real-time performance of the traditional target detection algorithm and solve the problems of poor effect and high miss rate in small target detection,an improved Center-Net algorithm was proposed.Firstly,the feature extraction network was changed from ResNet50 to SqueezeNet.The convolution calculation parts in the network were replaced by depthwise separable convolution.Then,the double-threshold improved non-maximum suppression(NMS)algorithm was used to replace the single-threshold NMS algorithm,and the loss function was calculated through DIoU.The experimental results show that the detection accuracy of the improved algorithm in helmet detection and mask detection datasets is 91.3%and 85.5%.Compared with the original CenterNet algorithm,the performance is improved by 2.35%and 3.76%,respectively,and the detection speed is faster.
万方数据
维普
