Research On Underwater Target Detection Algorithm Based On Deep Neural Network YOLO

In recent years,with the continuous growth of population and the continuous development of the international situation,the ocean has become the focus of national strategic development.The research on related fields related to the ocean is of great significance for national defense and security,the development of underwater fossil energy,marine environmental testing,and seafood fishing.Object detection is an important branch in the field of computer vision,widely used in projects such as autonomous driving,industrial detection,and real-time monitoring.Through object detection algorithms,costs can be significantly saved and production efficiency can be improved.Underwater target detection plays a crucial role in both military and civilian applications.With the continuous exploration of the ocean by humans,underwater target detection technology has gradually become a research focus in the field of underwater detection.The current underwater target detection model lacks sufficient feature extraction for images,and most targets are small.Some classic target detection algorithms currently have good detection results for large targets,but poor detection results for small targets,making them unsuitable for detecting underwater targets.To solve the above problems,an underwater target detection model that improves the FPN network and the fusion attention mechanism is proposed.The main work content of the thesis is as follows:(1)To address the issue of insufficient extraction of target features in the backbone network,a global attention mechanism has been introduced into the backbone network.Firstly,compare multiple attention mechanisms and choose to embed the global attention mechanism into the YOLOv5 s backbone network to enrich the expression ability of feature maps,enhance key feature extraction,and ignore redundant features,thereby improving the recognition ability of object detection.(2)In response to the problem of low accuracy in small target detection,bidirectional feature fusion algorithm and adaptive spatial feature fusion algorithm have been introduced in YOLOv5 s.In this underwater data set,the object to be detected is a small target object,and the bottom Receptive field of the image is large,which contains more small target feature information.With the deepening of the neural network,the feature information obtained becomes more abstract,ignoring the bottom information of the image,resulting in missing and wrong detection of small target objects.Bidirectional feature fusion algorithm and adaptive spatial feature fusion algorithm fuse low-level information and high-level abstract Semantic information of the image,improving the accuracy of small target detection in underwater data set.(3)A system for object detection was designed to address the difficulties of model deployment.When deploying object detection models on devices,there is often a problem of slow recognition speed due to the large weight file of the model,which occupies a large amount of computing space on the device.The computer based underwater target detection system designed in this article based on the Windows system can directly load models into neural networks,and use portable personal laptops to quickly perform target recognition.Design the main function interface through Py Qt5,add modules such as image loading,weight loading,and image detection to complete the detection of underwater targets.The experimental results show that the proposed YOLOv5 s based improved underwater object detection algorithm improves the detection accuracy of underwater image datasets.The underwater object detection system designed based on Windows system simplifies user operations and improves the efficiency of object detection.