研究改进神经网络的船舶红外图像边缘检测方法，提升边缘检测抗噪声干扰能力。采用块匹配的主成分分析方法对船舶红外图像实施去噪处理后，经梯度算子将降噪后船舶红外图像转换为二值图像；以BP神经网络为基础，通过附加动量法-自适应学习速率调整BP神经网络权值，提高网络训练鲁棒性；将转换后舰船二值图像作为改进神经网络的输入，在实施网络训练后得出输出值，依据输出值和设置阈值的对比结果，获取船舶红外图像边缘点，实现船舶红外图像边缘检测。实验结果表明：该方法降噪后船舶红外图像的PSNR值全部高于40 dB，降噪效果较好；可有效提取船舶红外图像边缘特征且边缘检测结果清晰、连贯，能够达到船舶红外图像边缘检测标准。
The edge detection method of ship infrared image based on improved neural network is studied to improve the anti-noise ability of edge detection. After de-noising the ship infrared image using the block matching principal component analysis method, the de-noised ship infrared image is converted into binary image by gradient operator. Based on BP neural network, the weight of BP neural network is adjusted by the additional momentum method - adaptive learning rate to improve the robustness of network training. The converted ship binary image is taken as the input of the improved neural network, and the output value is obtained after the implementation of network training. According to the comparison results of the output value and the set threshold value, the edge points of the ship infrared image are obtained to achieve the edge detection of the ship infrared image. The experimental results show that the PSNR value of the ship infrared image after noise reduction is all higher than 40 dB, and the noise reduction effect is good. It can effectively extract the edge features of ship infrared image, and the edge detection results are clear, high and consistent, which can meet the ship infrared image edge detection standards.
2023,45(7): 166-169 收稿日期：2022-10-09
DOI：10.3404/j.issn.1672-7649.2023.07.032
分类号：TP391
基金项目：四川省科技计划项目（23NSFSC1129）
作者简介：刘志东(1982-),男,硕士,副教授,主要研究方向为图像处理及软件工程
