Comparative Analysis of Different Deep Learning Methods to Detect Eight Major Mango-Leaf Diseases

Abstract

Meeting nutritional demands and helping to alleviate the global food problem are both greatly impacted by fruit production. Tropical Indian weather is ripe for plant diseases, which significantly reduce crop yields and cost farmers a lot of money. To put it in perspective, a healthy harvest depends on prompt diagnosis of plant diseases. This study develops different methods like InceptionV3, ResNet50, VGG16, VGG19 and Proposed method based on convolutional neural networks (CNNs) to identify eight of the most prevalent mango leaf diseases by analysing leaf photos. An innovative collection of region-specific photos was used to train this model specifically for the pattern of mango leaf diseases (Anthracnose, Bacterial canker, Cutting weevil, Die Back, Gallmidge, Powdery meldow, Reddust, and Soothy mould) in India. It is capable of classifying almost all mango diseases that are widely accessible.  The performance of developed models is evaluated with average accuracy & model loss and found to be 46.87% & 1.564%, 74.44% & 0.687%, 17.18% & 2.004%, 86.53% & 0.355%, and 90.74% & 0.14% respectively. In thirty iterations, the average accuracy of proposed model is found to be higher (90.74 %) and its model loss is found to be very low (0.14 %). Thus, the proposed model may aid in the early diagnosis of mango leaf diseases, which in turn can increase mango output and boost the national economy.