Improved Infrared Image Edge Detection Algorithm Based on DexiNed

Compared with optical image edge detection, there are fewer studies on infrared image edge detection, and most of them are based on traditional methods, such as edge detection operators and mathematical morphology. In essence, they only consider the sharp local changes of infrared images to detect edges, so they are always limited by low-level features. In this paper, an infrared image edge detection algorithm based on deep learning is proposed. Based on the dense extreme inception network for edge detection (DexiNed), the network capacity is reduced by removing the last main block, the image level difference is introduced into the loss function, and the parameters of the loss function are carefully set to optimize the network performance. In addition, by adjusting the natural image edge detection dataset to approximate the infrared image edge detection dataset, the improved model was trained to enhance the edge detection ability. The qualitative evaluation results show that the edge of the infrared image extracted by our method is accurate, precise, rich in detail, and fits human vision. A quantitative evaluation using the structural similarity indexmatrix (SSIM) and feature similarity indexmatrix (FSIM) indexes further shows the advantages of our method compared with other existing methods.