Efficient crop–weed semantic segmentation via knowledge distillation from DeepLabV3+ to Fast-SCNN

Precision agriculture plays a crucial role in sustainable food production and resource-efficient land management. A key component is the semantic segmentation of crops and weeds, enabling autonomous systems to minimize herbicide use and environmental impacts through targeted weeding. However, state-of-the-art segmentation models, such as DeepLabV3+, demand high computational resources, making them impractical for low-power agricultural devices deployed in the field. This study introduces a green artificial intelligence approach using a novel knowledge distillation (KD) framework that transfers semantic knowledge from a DeepLabV3+ teacher to a lightweight Fast-SCNN student model. The proposed approach was designed specifically for energy-efficient crop–weed segmentation on the CWFID dataset. The framework employs dynamic alpha scheduling to balance hard and soft label supervision and a patch-based training strategy to handle high-resolution field imagery efficiently. The distilled Fast-SCNN achieves a mean Intersection over Union (mIoU) of 0.8695 and a pixel accuracy of 0.9879, with a compact model size of only 4.54 MB and real-time inference capability (4.84 FPS on an NVIDIA T4 GPU). Compared with several recent lightweight architectures, our method achieved competitive accuracy with significantly reduced computational and energy costs. The findings demonstrate that knowledge distillation can support sustainable AI practices by reducing the carbon footprint of deep learning models while maintaining high performance in precision agriculture applications. • A green AI framework enables energy-efficient crop–weed segmentation. • Knowledge distillation transfers DeepLabV3+ features to lightweight Fast-SCNN. • The distilled model achieves 0.9879 accuracy with only 4.54 MB parameters. • The approach reduces energy use while maintaining high segmentation quality.