Shun Ren, Xuan Liu, Hongyan Liu, and Lu Wang
Attention mechanism, residual network, PSPNet, cultivated land, remote sensing image
The monitoring of cultivated land area is an important content to ensure food security, among which the real-time monitoring of cultivated land using remote sensing image is one of the most important means to guarantee the same. Remote sensing image from the Gaofen-2 satellite is often affected by seasonal climate in the process of obtaining the information of topographical features, which makes the colour contrast of the data decrease and leads to classification errors. Therefore, the cultivated land images in spring and winter are taken as the experimental objects so as to address the problems of “the same spectrum foreign matter” between spring cultivated land and vegetation, winter cultivated land and bare soil. Paper contents are as follows: (1) The Pyramid Scene Parsing Network with residual attention mechanism is used to extract cultivated land; and the residual attention mechanism is applied to automatically adjust the weight to enhance the useful features and improve the segmentation effect; (2) to reduce the impact of semantic information loss on edge detail recognition with the deepening of network layer, the method of high-resolution representation of detail features in shallow attention network layer is integrated to optimize the network; (3) because of the unbalanced proportion of samples in the data set, the network training is easy to over fit and the dice loss function is used to solve the problem of imbalance between positive and negative samples. It can be seen from the experimental results that compared with the classic algorithm, under the influence of spring and winter, the cultivated land identification effect will be better and the application of farmland extraction can be realized.
Important Links:
Go Back
