A method of photoelectric ship image detection based on improved SSD
zhang mengyu, liu ningbo, wang zhongxun, yu hengli · 2023
There are inevitable problems in the imaging mechanism of visible and infrared images, which bring more difficulties to the detection task. Firstly, this paper introduces the principle of SSD one-stage detection model. Aiming at the characteristics and problems of the model, residual network is used to improve it, and the prediction feature layer is reselected and sorted. Based on the photoelectric equipment, the video data of marine ship targets are collected and the data set is constructed. Through the comparison of the training results of various models, it is found that the detection accuracy of small-sized targets is low. In order to solve this problem, First, reduce the size of the anchor , and the feature extraction network is modified. The experimental results show that the average detection accuracy of the improved model increases by 7%, and the single category accuracy of small targets increases to 87%.