Abstract: Crop semantic segmentation can identify different crop regions in images, providing pixel level fine-grained information that helps agricultural researchers better understand the distribution and status of agricultural resources, thereby supporting more accurate agricultural management and decision-making processes. This article proposes a new benchmark dataset for semantic segmentation of agricultural objects, (Shawan Crop Classification Dataset, SWCC). The data is collected from the Shawan area of Xinjiang, China, captured by Gaofen-1 satellite, and includes data with a resolution of 2 m panchromatic and 8 m multispectral. After fusion, 2 m multispectral data is obtained. All objects in the SWCC dataset are annotated with 5 categories through precise visual interpretation and field investigation. Compared with existing specialized agricultural object semantic segmentation datasets, the SWCC dataset has three characteristics. Firstly, it covers diverse crop categories in Xinjiang and has high representativeness, providing valuable data resources for agricultural scientific research. Secondly, achieve 2 m high-resolution multispectral data fusion, significantly improving image details and information content. Thirdly, select the RGB and near-infrared bands to comprehensively extract crop feature information. Based on the SWCC dataset, this article evaluates several state-of-the-art algorithms, providing a benchmark for deep learning based crop semantic segmentation methods, which is valuable for improving evaluation algorithms.