基于深度学习的卫星遥感图像边缘检测方法

为解决卫星遥感图像边缘模糊噪点过多，导致图像清晰度过低的问题，提出基于深度学习的卫星遥感图像边缘检测方法；利用Softmax分类器结构，提取边缘图像节点处的数据信息参量，遵循深度学习算法完成对图像信息的卷积与池化处理，基于CV算法实现基于深度学习的卫星遥感图像识别；根据尺度空间定义原则，确定边缘检测特征点所处位置，再联合梯度信息熵计算结果，完成对卫星遥感图像的拼接处理；分别计算一阶微分边缘算子、二阶微分边缘算子的具体数值，确定梯度幅值的取值区间，总结已知数值参量，建立完整的双阈值表达式，完成基于深度学习的卫星遥感图像边缘检测方法的设计；实验结果表明，应用所提方法后卫星遥感图像边缘节点处信噪比指标在55.1～62.7 dB范围内，图像模糊噪点个数最大为1.32×10⁵个，可获得较为清晰的遥感图像，在卫星遥感图像边缘精准检测方面具有较强的实用性。

An edge detection method of satellite remote sensing image based on deep learning

In order to solve the problem of too much blurred noise at the edge of satellite remote sensing image, which leads to low image definition, an edge detection method of satellite remote sensing image based on deep learning is proposed. Using the Softmax classifier structure, the data information parameters at the edge image nodes are extracted, and the deep learning algorithm is followed to complete the convolution and pooling of image information, and realize the recognition of satellite remote sensing images based on deep learning. According to the definition principle of scale space, the location of edge detection feature points is determined, and the result of gradient information entropy calculation is combined to complete the splicing of satellite remote sensing images. Calculate the specific values ??of the first-order differential edge operator and the second-order differential edge operator respectively, determine the value range of the gradient amplitude, summarize the known numerical parameters, establish a complete double-threshold expression, and complete the satellite remote sensing image based on deep learning. Design of edge detection methods. The experimental results show that the signal-to-noise ratio index at the edge nodes of satellite remote sensing images increases after the application of the proposed method, which can effectively control the impact of blurred noise on image clarity, and has strong practicability in accurate edge detection of satellite remote sensing images.