This study is an application of computer vision techniques to the extraction of the boundaries of rice fields in the middle valley of the Senegal River. The objective of this work is to develop a method for an automatic and reliable parcelization of the contours of irrigated rice plots to practice Sentinel2 images. The automatic delimitation of plots is a new field of research to allow managers of rice fields in the Senegal River valley to have the exact areas of the sowings. It is based on Sentinel 2 data and semantic segmentation methods using neural networks. The methodological approach consists of using a convolutional neural network to detect parcel boundaries. The results obtained from the extraction of the contours remain entirely satisfactory for a correct calculation of the yields per season. This technique of automation and calculation of rice areas makes it even more reliable and sustainable for updating agricultural databases.

Keywords : Sentinel 2, Segmentation, Plot, Irrigated Rice Field, Neural Network.