Segment Anything Model (SAM) Applications: Coffee Rust Detection with Image Processing

Coffee Leaf Rust (CLR) poses a significant threat to coffee production worldwide, with the potential to cause substantial yield reductions in severe cases. Early detection of such diseases is crucial to mitigate severe damage and economic losses. Addressing this issue requires precise and efficient detection methods. This paper explores the development of two algorithms for detecting CLR in coffee leaves using the RoCoLe and BRACOL datasets. The first algorithm, termed exclusive SAM, relies solely on the Segment Anything Model (SAM), while the second algorithm, termed SAM Assisted, uses SAM in conjunction with additional image processing techniques. The results demonstrate that the exclusive SAM algorithm achieves an accuracy of 92% on the RoCoLe dataset and 88% on the BRACOL dataset. In contrast, the SAM Assisted algorithm achieves an accuracy of 94% with the RoCoLe dataset and 96% with the BRACOL dataset. Lastly, this research identified the difficulties the Segment Anything Model (SAM) faces when performing specialized segmentation tasks on its own. However, it also highlighted SAM's potential as a supporting tool for more robust algorithms. In the future, this will likely be the most important role of SAM in precision agriculture research. Additionally, the second algorithm developed using SAM showed very promising results with image techniques not conventionally used in agricultural applications, warranting further investigation.