Marine ship detection method based on improved YOLOV3 algorithm

Abstract

The invention belongs to the field of computer vision target detection, and particularly relates to a marine ship detection method based on an improved YOLOV3 algorithm, which can be used for marine ship target detection and comprises the following steps: step 1, designing a network structure with multi-scale feature fusion; step 2, designing a feature information interaction network structure; step 3, optimizing a model loss function; step 4, carrying out data set balancing processing and priori frame clustering; step 5, training a model training, wherein the model comprises a data preprocessing module and an improved YOLOV3 network structure; and step 6, using the trained model to predict the type and position information of the ship for the target image. Compared with an existing detection algorithm based on deep learning, the method invention has high detection precision and can be suitable for ship detection in shot color images.

Description

technical field [0001] The invention belongs to the field of computer vision target detection, in particular to a marine ship detection method based on the improved YOLOV3 algorithm, which can be used for marine ship target detection. Background technique [0002] Aiming at the problem of ship detection and recognition under the ocean background, there are currently two common solutions. One is the imaging series, which is mainly based on the detection methods of SAR images and infrared images based on aperture synthetic radar. Its application is relatively mature, but its Imaging fails to contain rich spectral information of the target, and deviates from human visual habits, which is not conducive to the network to extract richer feature information, thereby reducing the accuracy of detection. Another type of non-imaging series uses sonar and other technologies to obtain relevant signals of ship targets, but the maritime environment is complex, and the communication transmi...

AI Technical Summary

Problems solved by technology

However, when using neural networks for image detection, the biggest problem is the lack of data sets, and the second is that in the face of low-quality pictures, if the network learning ability is not enough, the target cannot be detected, and false detections and missed detections are more likely to occur.

Images