In the process of power inspection, a large number of images are generated, and the original image files are not suitable for transmission and storage. In this context, a hybrid compression method for power equipment inspection images is proposed based on the RetinaNet target detection algorithm and region adaptation. Initially, the RetinaNet algorithm is employed to identify objects in the image, with detection performance optimized through Mosaic data enhancement and adaptive anchor box screening. Subsequently, utilizing the detected target area, a region-adaptive compression strategy is applied to achieve end-to-end compression based on convolutional neural network for the target power system equipment; meanwhile JPEG2000 image compression is utilized for background images to ensure high compression efficiency without compromising image quality. Experimental results from multiple datasets demonstrate that the proposed method has led to improvements in object detection accuracy and image compression efficiency.