Research On Ship Target Detection Algorithm Based On Feature Fusion

As maritime area supervision and shipping management play an important role in promoting national security and economic development,the development of ship detection technology has received more and more attention.Traditional target detection methods require human design methods,consume a lot of time and labor costs,and the recognition accuracy needs to be improved.Therefore,the establishment of a ship target detection method based on deep learning is the focus of this article.Due to complex sea conditions,such as light reflection on the sea level,weather conditions,waves and other factors will cause interference to the detection results of ship targets.Therefore,based on the YOLOV4 target detection algorithm,which has excellent performance in feature fusion,this thesis combines Focus and Convolutional Block Attention Module(CBAM)to construct a ship target detection algorithm based on attention mechanism(Focus and CBAM You Only Look Once model,FC-YOLO).It not only solves the problem that small target ships are not easy to detect,but also increases the ability of the model to extract features,making the algorithm more adaptable to the interference of complex sea conditions and improving the model detection accuracy.In addition,this thesis classifies and filters the publicly available optical remote sensing image datasets.However,it is difficult to meet the needs of today’s tasks only by identifying ship targets.Therefore,this thesis constructs a "yacht-cargo ship" ship type optical remote sensing data set,and trains the algorithm to make it able to have the ability to identify the type of ship.Comparative experiments show that the constructed FC-YOLO algorithm is not only more accurate in prediction effect,but also has good target detection speed.This thesis also uses training warm-up and mosaic data enhancement methods in the training process,which improves the generalization ability of the algorithm and further optimizes the performance.Due to the complexity of the deep learning algorithm,the algorithm not only consumes a lot of time and resources in the training phase,but also has time loss in the test phase.For the task of ship target detection,it is very important to provide timely and effective information for shipping management through real-time detection.Therefore,this thesis proposes to build a ship target detection algorithm based on lightweight feature fusion network on the basis of FC-YOLO,and introduces the lightweight feature extraction network into the ship detection algorithm,saving the time needed for feature extraction,which is the most time-consuming in the training process.Experimental verification shows that compared with the target detection network before the improvement,the ship target detection algorithm based on the lightweight feature fusion network still has considerable recognition and detection accuracy,and it handles light and shadow,sea waves,and occlusion well.At the same time,the detection speed has been significantly improved,which makes it more advantageous in the real-time detection of ship targets.