Small object target detection method and system of lightweight multi-scale attention mechanism

The invention provides a small object target detection method and system based on a lightweight multi-scale attention mechanism. The method comprises the following steps: step 1, extracting features by using GhostNet as a backbone feature extraction network of a YOLOv4 target detection architecture; step 2, for the features extracted in the step 1, using a multi-scale attention module to capture features with discrimination in a small target image from two dimensions of space and channel; and step 3, adopting a Soft-NMS algorithm to reduce the confidence coefficient of a detection frame overlapped with the current optimal detection frame for the feature map output in the step 2. The method is small in network structure size, high in detection speed and good in small target detection effect, completely meets the requirement of a real-time scene, and has very high practical value. 本发明提供一种轻量级多尺度注意力机制的小物体目标检测方法及系统，方法包括如下步骤：步骤1，利用GhostNet作为YOLOv4目标检测架构的主干特征提取网络提取特征；步骤2，对步骤1所提取到的特征使用多尺度注意力模块捕获从空间和通道两个维度上对小目标图像中具有鉴别