Spatial-DeepLabV3+: Crop Segmentation in Satellite Images

Crop land segmentation is a challenging task in precision agriculture, focusing on identifying suitable crop types from satellite imagery of agricultural land. Existing machine learning and deep learning solutions for crop segmentation struggle with processing multi-spectral data, focusing relevant parts of image and handling illumination effects. To address these challenges and accurately segment crops in land images, we propose a deep learning model that processes images across multiple scales and spectral bands, using atrous convolutions to effectively capture local features. Additionally, the model incorporates spatial attention to emphasize informative regions and suppress irrelevant features, enhancing segmentation accuracy for specific land types. We evaluated the model on the DeepGlobe Land Cover dataset, and the results demonstrate that our approach outperforms existing crop segmentation solutions in terms of accuracy, precision, and recall.