关键词: 农业遥感, 蔬菜大棚, 多时相, 基于补丁长短时记忆网络, 语义分割模型, 卷积神经网络

Abstract:

Greenhouse is of great significance to agricultural production. Due to the influence of seasons and environment, the features of remote sensing images in different periods are different, and the accuracy of single-temporal features can not meet the requirements. In recent years, deep learning has been proved to be suitable for the classification of remote sensing data. In order to realize the effective application of deep learning in agricultural remote sensing, this paper designs an improved multi-temporal semantic segmentation model called Multi-temporal Spatial Segmentation Network（MSSN） for greenhouse extraction. The Patch-LSTM（based on Patch long time-memory network） is proposed, which makes full use of the spatial and temporal information of images. Spatial pyramid pooling with void convolution （ASSP） is used to solve the problem of network sensitivity to scale. The skip-layer and deconvolution layer are further added to improve the reduction ability of the feature map Shandong Gaomi is chosen as the experimental area, and the multi temporal GF-2 remote sensing image is used for the experiment. The results show that the segmentation model has 0.95 precision, 0.92 F1 score and 0.93 foreground IOU（Intersection Over Union） on the test set. The segmentation model can achieve high-precision greenhouse extraction and provide a new method for the application of deep learning in agricultural remote sensing.

Key words: agricultural remote sensing, greenhouses, multi-temporal, Patch-LSTM, semantic segmentation model, Convolutional Neural Network（CNN）