This research aimed to extract agricultural land use pattern around the Miyajimanuma wetland, Hokkaido, Japan. By combining the image segmentation technology - watershed transform and image classification technology- particle swarm optimization (PSO)-k-means based minimum distance classifier, a new method for extracting the agricultural land use information based on remote sensing imagery was developed. Remote sensing image segmentation and classification provide a quick method for estimating important crop characteristics or locations of various crops in a field that appears to have similar characteristics and knowing the impact of agricultural production activities to the environment. Watershed transform algorithm was used to extract the land parcels in the catchment; PSO-k-means based minimum distance classifier was used to do remote sensing imagery classification which classifies an agricultural land into five classes including paddy, soybean field, wheat field, water body and others. Finally, since crops planted in a certain field are usually the same, the pixels in one parcel were divided into one class according to the result of classification, although some noise pixels may appear to be in another class, and thereby extracting the land use pattern. Results from the study indicated that by using this method the accuracy of the crop classes identified was over 96%. As such, the proposed method could be of interest to researchers who need more accurate information on agricultural land use.