A method for radar time-frequency image data detection based on an improved YoloV4 network

In sea surface target detection missions, due to the variable weather conditions at sea, the complex and variable surges formed in adverse weather conditions make target detection difficult, and radar sensors are widely used based on their ability to reliably detect targets in all adverse weather conditions. Sea clutter is defined as the backscattered echoes formed by radar irradiation onto the sea surface. Detection of small targets on the sea surface in the background of sea clutter is usually faced with unfavourable factors such as low signal-to-noise ratio. This thesis proposes a method for radar time-frequency image data detection based on an improved yolov4 network, which can perform the task of sea surface target detection better and can achieve an accuracy of 90.11% on the measured radar time-frequency image dataset, outperforming similar classical deep learning methods, and ablation experiments are done to demonstrate the effectiveness of the improvement.