微型电脑应用

由于无人机巡检图像边缘检测的距离误差大、图像清晰度低，提出基于全卷积神经网络的无人机巡检图像边缘检测方法。采用水平集量化特征分解方法，提取无人机巡检所采集图像的多尺度边缘特征；采用全卷积神经网络构建图像边缘检测模型结构，优化损失函数，完成模型的局部和整体训练，并将多尺度边缘特征输入深度学习网络；采用二阶导数计算像素边缘概率，检测图像的弱边缘并生成边缘信息概率图，计算无人机巡检图像弱边缘对象的概率值，实现图像边缘细化。实验结果表明，所提方法能有效获取图像中目标对象的边缘特征，距离误差均小于0.25,图像清晰度均在24以上，能够完整、可靠获取图像中不同位置、物体等目标的边缘结果，且边缘检测结果更为精细。

Due to big distance error and low image definition of UAV inspection image edge detection, a UAV inspection image edge detection method based on full convolution neural network is proposed. Inspection images, multi-scale edge features of UAV inspection images are extracted by level set quantization feature decomposition method. The full convolution neural network is used to construct the image edge detection model structure, optimize the loss function, the local and global training of the model is completed, and multi-scale edge features are inputted for deep learning. The second derivative is used to calculate the pixel edge probability, detect the weak edge of the image and generate the probability graph of edge information, calculate the probability value of the weak edge object of the UAV inspection image, and realize the image edge refinement. The experimental results show that the method can obtain the edge features of the target object in the image effectively, the distance error is less than 0.25, and the image definition are all above 24, which can obtain the edge results of different positions and objects in the image completely and reliably, and the edge detection results are more refined.