| The target characteristic data of maritime ships has an important supporting role for the development and use of weapons and equipment.Since there are many maritime ship targets,it is important to study maritime ship target detection and classification technology to filter out specific targets and support maritime ship target characteristics data collection.Aiming at the joint radar and optoelectronic equipment detection scenarios of maritime targets,this paper focuses on the automatic labeling method of maritime ship target data,the integrated processing method of radar image layer clutter suppression and ship target detection,and the classification method of ship target detection in optoelectronic equipment images to support the long-term accumulation and application of specific maritime ship target characteristic data.The main work of this paper is as follows:(1)To address the problem of sea state level information affecting the selection of target detection method,the characteristics of sea clutter raw data are analyzed.Three radar time domain echo sequences are converted to the frequency domain,and the classification method of sea state level is studied on the basis of the frequency domain.It is found that Le Net achieves a better classification level for sea state echo sequences under the premise that the training test data have similarity,while the classification effect of sea state class is not satisfactory under the condition that the data type are different.(2)A Yolov5 clutter suppression feature enhancement network based on automatic annotation is proposed to address the problems of time-consuming,labor-intensive and biased annotation of dense ship targets and small-size ship targets,as well as the interference of target pixel values and sea clutter in the detection of ship targets at sea.Firstly,a suitable CA-CFAR detector is selected according to the sea state characteristics to achieve automatic labeling of ship target positions,secondly,a deep learning method combining feature enhancement and target detection is established,and clutter suppression is achieved by the method of contextual information fusion plus attention mechanism,and ship target detection is accomplished by using the improved prediction output layer,and finally,the automatic labeling is verified using the measured data Finally,the effectiveness of the proposed method is verified by using the measured data.Finally,the automatic labeling and the effectiveness of the proposed method are validated using the measured data.The validation of the measured data shows that the automatic labeling data meet the requirements of the network training and the proposed method achieves a recall rate of 97.96% based on the Pdisplay image.(3)In response to the problem that the detection of ship targets in radar images is insufficient for the classification and recognition of radar ship targets,the photoelectric image data of ship targets intercepted by the guidance of radar to photoelectricity are combined to further realize the detection and classification of ship targets at sea on this basis.Firstly,an improved version of SSD network for small targets of ships is proposed for the problem that the detection effect is not satisfactory due to the inconspicuous features of small targets in the photoelectric ship images.Secondly,the phenomenon of poor detection performance due to the target itself and the network framework is analyzed by the constructed multi-category data set.Finally,the effect of the adjustment of the candidate frame size and the strategies of inserting the residual module and adjusting the prediction feature layer in the backbone network on the detection results of small ship targets is investigated,and the superiority of the proposed method for improving the classification performance of ship target detection is verified using the measured data.The validation of the measured data shows that the detection accuracy of the proposed method for small ship targets is improved by 7%,and the robustness of the network still does not tend to decrease when the samples are not sufficient. |
