Remote Sensing Image Object Detection Based on Improved YOLOv7 Algorithm

Remote sensing object detection is to use remote sensing image data acquired by remote sensing technology to automatically detect and identify target objects in images by computer algorithms. An improved yolov7 remote sensing image object detection algorithm is proposed for the problems of low detection accuracy and slow detection speed of traditional remote sensing image object detection algorithms. The SPD (Space-to-depth) convolutional building block is introduced to improve the detection performance of small and medium-sized targets in remote sensing images at the output; meanwhile, the ECA (Efficient Channel Attention) module is added to the Neck structure to improve the feature fusion capability of the network. The ablation and comparison experiments are done on DIOR dataset, and the results show that the improved YOLOv7 algorithm has a mAP value of 82%, which is 6.8 percentage points higher than the original YOLOv7 algorithm. The average detection time of single image is 13.8 ms, which is basically the same as the original algorithm, and the detection performance is better than that of the faster-rcnn and ssd algorithms, which can effectively detect remote sensing targets.