부산대학교 도서관

This study is about highlighting the identification and detection of guava fruit and guava leaves disease using some modern techniques like Convolutional Neural Networks i.e., CNN, Artificial Intelligence i.e., AI, Deep Learning, Random Forest, and some models like GIP-MU-NET, DarkNet-53, AlexNet, Sidney. The methodology consists of 4 parts: data collection, data processing, data extraction, and classification respectively. The total data that comes in the hybrid approach category are collected from primary sources, which means data are collected from self, and secondary sources, which means data are collected from other sources such as Punjab, Himachal, and Haryana. The research includes 7 classes. The maximum precision acquired in Class-4 is (93.52%) while the minimum precision in Class-7 is (65.72%). The macro average for precision, recall and F1-Score is 77.08, 75.94, and 76.26 respectively. The weighted average for precision, recall and F1-Score is 77.61, 76.78, and 76.92 respectively. The Micro Average for precision, recall, and F1-Score is 76.78, 76.78, and 76.78 respectively. The early identification of the disease of the fruit helps to survive them from diseases like wilt, die back and pepper spot, stem corruption and dry fruit rot, citrus canker, dulse flag, crop mark, and styler end rot ultimately it also helps to improve economic growth. There are certain limitations also like Attention-based model can be used in the research, also images used can be more, GANs can be used for data augmentation, and more varieties can be included.. In the future, this study will help to increase productivity by identifying the affected fruit to farmers.