Research on Ship Target Detection Based on Deep Learning and Image Segmentation
A new ship target detection method is proposed by integrating deep learning techniques,convolutional neural networks,and differentiable binarization techniques to address the issues of target scale changes,complex shapes,and background interference in the field of ship target detection.This method adopts an encoding decoding structure for ship target detection network,introducing deformable convolution and differentiable binarization functions to improve adaptability and detection performance.In network architecture,the design of embedding deformable convolutions helps to more effectively capture the features of ship targets,adapting to the variability of target scale and shape in complex scenes.The differentiable binarization technique can retain accurate information about the target boundary and improve sensitivity to the shape and contour of the target.During the training process,the supervised training threshold graph method was adopted,and the adaptive threshold for each position was determined through joint optimization,reducing the complexity of post-processing steps.The detection accuracy of the proposed method was verified through detection experiments on the HRSC2016 and DOTA datasets.
deep learningCNNship target detectiondeformable convolution
NETL
NSTL
万方数据
