基于YOLOv5 改进的小目标检测算法

针对传统目标检测算法存在对小目标检测的识别精度低和不稳定的问题，提出基于YOLOv5 改进的小目 标检测算法。基于卷积神经网络加入额外的检测头，采用数据增强策略并更改网络卷积步长，解决了小目标像素低、 占比小、易重叠和难以分辨等问题；同时依托真实检测场景制作一个全新的针对飞机检测的卫星影像数据集，该数 据集的待检测小目标占比达61%，飞机姿态及场景丰富，有助于客观全面地验证网络精度。将改进后的算法与原始 的YOLOv5 模型进行对比，结果表明，其平均精确率AP 值较原始YOLOv5 模型提升约3%。

Improved Small Target Detection Algorithm Based on YOLOv5

Aiming at the problem of low recognition accuracy and instability of small target detection in traditional target detection algorithm, an improved small target detection algorithm based on YOLOv5 is proposed. Based on the convolutional neural network, an additional detector is added, the data enhancement strategy is adopted and the network convolution step is changed to solve the problems of low pixel, small proportion, easy overlap and difficult resolution of small targets. At the same time, relying on the real detection scene, a new satellite image data set for aircraft detection is produced, in which the proportion of small targets to be detected is 61%, and the aircraft attitude and scene are rich, which is helpful to verify the network accuracy objectively and comprehensively. Comparing the improved algorithm with the original YOLOv5 model, the results show that the average accuracy AP value of the improved algorithm is about 3% higher than that of the original YOLOv5 model.