Rice shrimp co-cultivation, as a circular agriculture model that integrates ecological, economic, and social benefits, has been widely applied in China. Accurately obtaining information on the planting area, spatial distribution, and changes of rice shrimp co-cultivation is of great significance for ensuring safe production, social supply of rice and shrimp, and formulating corresponding management and planning policies by government departments. Most of the existing methods for extracting rice shrimp farming fields used medium-resolution temporal images as data sources to analyze the characteristics of water or vegetation changes within a year to construct models. The accuracy of the extraction depended on the results of water or vegetation analysis, the boundaries were unclear and there was a lot of noise. To solve the problems of complex temporal analysis, low accuracy of extraction results, and incomplete boundaries in existing extraction methods, this paper used high-resolution images of a single temporal phase as the data source, without analyzing the temporal features of rice shrimp farming fields. Only the spatial features of rice shrimp farming fields on high-resolution images were used for high-precision extraction. We proposed a deep learning semantic segmentation model named EASNet (edge assisted segmentation network), which was mainly composed of three modules: feature extraction module (FE Module), edge assist module (EA Module), and information fusion Module (IF Module). The FE Module was used to extract multi-level and fused contextual semantic information features of the target object. The EA module used the unique edge "shrimp groove" of rice and shrimp farming fields as an auxiliary information for separate segmentation in the designed EA Module. The IF Module integrated the outputs of the FE Module and the EA Module, enabling the main task to not only enhanced the boundary structure information of rice shrimp farming fields but also learned the unique spatial and semantic information of rice shrimp farming fields. The experimental results showed that with the enhancement of the EA Module, the segmentation results of rice shrimp farming fields were more complete and the boundaries were clearer. The IoU (Intersection over Union) of semantic accuracy and the F1-Score of boundary accuracy were improved by 1.5% and 5.8%, respectively. The recall, IoU, and F1-Score of overall semantic accuracy reached 0.970, 0.964, and 0.930, respectively. The Recall and F1-Score of boundary accuracy reached 0.864 and 0.859, while the Recall and F1-Score of relaxed boundary accuracy reached 0.876 and 0.913. The trained EASNet model was applied to the whole area of Xuyi County, and the spatial distribution map of rice shrimp farming fields in Xuyi County in 2020 was obtained. In comparison with the results of rice shrimp breeding fields extracted by the traditional Water Seasonal Difference method and the Random Forest method, our method obtained the optimal results with an OA(overall accuracy) of 96.71% and a Kappa coefficient of 0.934. The method used in this paper has higher accuracy, lower missing rate, more regular boundaries, and fewer instances of similar integration in extracting results. It can provide a basis for natural resource surveys, and government departments to formulate corresponding rice shrimp breeding management and planning policies.

High spatial resolution of remote sensing imagery has been increasing with the optical sensor performance. An accurate and rapid classification of the crops is often required for agricultural production, yield prediction, and structure adjustment. However, the traditional high-resolution imagery cannot fully meet the rich phenological information in the crop growth cycle, particularly in the complex planting structure regions with both single- and double-season crops. This limitation can significantly constrain the performance of the high-resolution crop semantic segmentation models. This study aims to propose a multi-source crop semantic segmentation model—MCSNet (multi-source crops segmentation network)—that integrates the high-resolution remote sensing imagery with the medium-resolution time-series normalized difference vegetation index (NDVI) data. A dual-encoder structure was composed of a high-resolution encoder (HR-Decoder) and a time-series encoder (TS-Decoder). The HR-Decoder was targeted at the high-resolution imagery in order to extract the spatial detail features, such as the crop plot boundaries and texture differences. Meanwhile, the TS-Decoder was focused on the medium-resolution time-series NDVI data. The vegetation indices were utilized to sensitively capture the spectral variations of crops throughout their growth cycles, thereby fully exploiting the phenological features to distinguish between single- and double-season crops. Furthermore, the network incorporated the convolutional long short-term memory (ConvLSTM) units within the TS-Decoder. The modeling capacity was enhanced for the complex temporal information. The local spatial features were extracted to effectively capture the dynamic changes along the temporal dimension. Subsequently, a multi-feature fusion encoder (MF-Encoder) was integrated to fuse the multi-source features from the HR-Decoder and TS-Decoder. The residual double attention was also utilized to emphasize the importance of the critical feature channels and spatial positions. Thereby, the temporal features and high-resolution spatial details were fused to ultimately strengthen the precision and robustness of the crop classification. The time-series NDVI data were also integrated with the MCSNet in the experimental phase. The accuracy of the crop classification was significantly improved, compared with the traditional only on the high-resolution imagery. The comparative experiments showed that the MCSNet shared the outstanding performance, with the mean intersection over union (mIoU) of 77.75% and an overall accuracy (OA) of 89.56%, indicating the highest levels. Furthermore, the ConvLSTM and the residual double attention in the network enhanced the modeling capability of the spatiotemporal features, thus increasing mIoU and OA by 3.84 and 4.24 percentage point, respectively. The MCSNet model was applied to the large-scale and complex study area of Xuyi County, Huaian City, Jiangsu Province, China. According to the pixel- and object-oriented classification, like Bi-LSTM (bidirectional long short-term memory), MCSNet exhibited significant advantages in both mapping and classification accuracy. Specifically, the MCSNet achieved an OA of 89%, a Kappa coefficient of 0.85, a mean weighted F1 score (mF1) of 0.89, and an mIoU of 0.78, thus outperforming the comparative data across all metrics. Therefore, there were the effectiveness and practicality of the MCSNet for the classification tasks in the large-scale and complex planting structure regions. In summary, the MCSNet can offer a viable technical pathway for multi-source data processing by integrating high-resolution imagery and time-series NDVI data. The dual-encoder structure (ConvLSTM) and residual double attention module (MCSNet) were introduced to effectively enhance the crop classification accuracy and stability in the complex planting structure regions. This finding can also provide a strong theoretical and technical solution to the multi-source remote sensing data fusion for crop production and structure optimization in sustainable agriculture.