UNDERSTANDING ARTIFICIAL INTELLIGENCE'S PROGRESS IN PLANT DISEASE DIAGNOSIS: A COMPARATIVE SURVEY ON COTTON PLANT

Abstract

Cross-domain transfer learning is becoming increasingly popular as an effective technique for disease detection in cotton crops with the help of artificial intelligence. Though the concept promises a lot, there are still a number of constraints and research loopholes that need to be filled. One big challenge is that many models are trained on clean, high-quality images from areas like medical scans or face detection, but real farm images are messy, natural, and often lower in quality. This mismatch makes it hard for the models to perform well in agriculture. Another challenge is the limited number of large labeled datasets. Particularly for cotton plant diseases, this hinders the ability to fine-tune or train models effectively. There are also no standardized benchmarks to assess and compare various models on an equitable basis. These gaps show why we need better models for agriculture, bigger datasets, and common testing methods, so AI tools can truly help farmers in the field. In this study, the performance of different YOLO versions was analyzed for cotton plant disease detection. Experimental results show that YOLOv8 achieved the best overall accuracy (95.14%) for cotton disease detection. The results provide valuable insights for researchers to advance further in this field and explore new opportunities.