Ship target detection algorithm based on improved deep learning

Abstract

Faster R-CNN algorithm is a deep learning network model based on regional suggestion network, which is applied to the field of target detection and recognition. Extracting a small number of pixels in the original image and generating a downscaled image can improve the detection efficiency of Faster R-CNN. Scene semantic narrowing is be directed at specific types of regions and geographical locations in the image. By performing targeted analysis on these regions, it can be interpreted more carefully, and the image is better applied to Faster R-CNN. The deep convolutional network with the narrowing function of the theme is studied. The main factors that affect the understanding of image content accumulated by human visual cognitive experience and computer vision research are summarized into various themes, and the appropriate theme is selected as a narrow sub-network according to the task. Realize the optimization of the Faster R-CNN algorithm by implementing a collaborative deep network with clear functions in the overall black box and subnet. The experimental results show that the proposed method can significantly shorten the detection time of the algorithm while improving the detection accuracy of Faster R-CNN algorithm.