Remote Sensing Image Extraction of Drones for Agricultural Applications

With the rapid development of national agriculture, the problem of large area occupied and insufficient image resolution has become increasingly prominent. In order to solve the problem of low resolution efficiency, this paper determines the optimal segmentation scale and constructs a multi-scale segmentation hierarchy network, and finally determines the three scale layers Level 60, Level 85 and Level 120 as the appropriate extraction scales of the corresponding feature types. The experimental results show that the segmentation object obtained by this method is close to the boundary of the actual object; this paper uses multi-scale segmentation to extract different scale entities as different objects, and makes full use of low-level or high-level object features, and can pass the mask relationship. The classification of features that are not easy to distinguish makes the image classification more flexible and closer to the human thinking process. Based on the segmentation and classification of the orthophotos combined with the topographic factors, the combination of the two types of data is sufficient. Using the spectral, texture, shape information and topographical feature information of the image, in the classification extraction based on object-oriented analysis and supervised learning; the research shows that the classification effect based on SVM is better than KNN algorithm.