Improved Mango Disease Detection Model for Energy-efficient Edge Computing Device

Mango is a fruit that has high economic value for several countries, so it is widely cultivated through technological engineering, including the implementation of artificial intelligence through vision sensing based on object detection methods.In this study, we proposed a detection model by integrating a You-Only-Look-Once version 7 Tiny (YOLOv7-Tiny) detection model and U-Net segmentation to detect plant growth through mango leaf diseases.The dataset development was carried out through image collection from several resources and image augmentation, which resulted in a total of 13004 images consisting of eight classes, namely, anthracnose, bacterial canker, cutting weevil, die back, gall midge, sooty mold, powdery mildew, and normal.The training, testing, and validation data were set to 70, 20, and 10%, respectively.The comparative experiment involved other models, namely, YOLOv4-Tiny, YOLOv5n, YOLOv7-Tiny, and YOLOv8n.The experimental results showed that the proposed model has outstanding performance compared with the other models with a mean average precision of 90.22%, while precision, recall, and F1-score have the percentages of 88.42, 87.93, and 88.17%, respectively.In practical applications, the proposed model has significant results in detecting mango leaf diseases.Moreover, the proposed model has good performance in terms of energy efficiency, which is represented by the model size and inference time generated by the model.