Research On Ship Target Detection Method Of Remote Sensing Image Based On Improved YOLOv5

Remote sensing image ship target detection is widely used in military and civilian fields,and is a key research problem in the field of remote sensing image processing.In recent years,with the rapid development of remote sensing technology,the acquisition of high-resolution optical remote sensing images has become easier,and these high-quality remote sensing images contain massive data information and rich detailed information,providing opportunities for the evolvement of optical remote sensing image object detection.Among them,ship target detection based on optical remote sensing images has been widely studied by scholars.However,due to a series of detection difficulties of ship targets,the existing target detection methods are difficult to achieve ideal results when applied to ship detection.Therefore,this paper studies ship target detection based on optical remote sensing images according to the characteristics of ship targets,in order to achieve fast and accurate ship target detection.The main work of this paper is as follows:(1)According to the characteristics of ship targets in remote sensing images,this paper collects remote sensing ship images in actual scenes,and manually annotates the images.The ship dataset A-Ship is produced and used for model training and testing.(2)Aiming at the problem of low detection accuracy due to complex ship target background,small target and multi-scale,a multi-scale ship target detection method based on self-attention mechanism is proposed.First,the self-attention mechanism is embedded in the backbone network to effectively locate the region of interest and enhance the network feature extraction ability.Then,through multi-scale detection,the shallow features are fully utilized to improve the detection performance of small objects.Finally,the improved bidirectional feature fusion module is used to enhance the sensitivity of the network to targets of different scales and improve the detection effect of multi-scale targets.The experimental results show that the method in this paper improves the detection accuracy of ship targets with complicated background,small targets and multi-scale characteristics.(3)Aiming at the problem of missed detection caused by the detection of ship targets in any direction and densely distributed by the horizontal frame,a ship target detection method based on rotating frame(RYOLOv5)is proposed.Firstly,the long-side representation of the circular smooth label is used to generate the rotating frame,and then the ship detection method of the rotating frame is realized through the rotation mosaic data enhancement,the number of network output channels,the loss function,and the tilted NMS.The experimental results show that the method in this paper improves the detection performance of ship targets in any direction and dense distribution,and also has a good detection effect on ship targets with complex backgrounds,small targets and multi-scale characteristics.(4)In order to better apply the method proposed in this paper to the actual scene,a set of remote sensing image ship target detection system based on browser and server architecture is designed and implemented.