Abstract: Greenhouse area statistics is the objective requirement of scientific management of vegetable industry. It can reduce labor costs better to acquire greenhouse area automatically through deep learning. In order to obtain the greenhouse area accurately, a framework based on deep learning is proposed to solve this problem. First of all, pyramid pooling module and chain pooling module are used to deal with high-level features of large scale and small scale respectively, and multi-scale semantic information is extracted under the condition of maintaining resolution and avoiding degradation. Then, the recursive residual module is used to alternately fuse the high-level and low-level1 information to further optimize the segmentation results. Comparing with three newer algorithms, the experimental results show that the proposed method has achieved the best results..