Analysis of small target detection algorithm based on SSD and YOLOv5
Wen Jing Zhou · Academic Journal of Computing & Information Science · 2023
SSD is a single-stage target detection algorithm, which performs feature extraction by convolutional neural network and takes different feature layers for detection output, so SSD is a multi-scale detection method. In the feature layer to be detected, a 3*3 convolution is directly used to perform the transformation of the channels. ssd uses an anchor strategy with pre-defined anchors of different aspect ratios, and each output feature layer predicts multiple detection frames (4 or 6) based on the anchor. A multi-scale detection approach is used, where a shallow layer is used to detect small targets and a deep layer is used to detect large targets. yolov5 is a single-stage target detection algorithm, which adds some new and improved ideas to yolov4, resulting in a significant performance improvement in both speed and accuracy. We conduct algorithm experiments with SSD and YOLOv5, and analyze the experiments to obtain better improvement ideas for small target algorithm.