Smart Horticulture Based on Image Processing: Guava Fruit Disease Identification

Guava fruit production is one of the main sources of economic growth in Asian countries; world production of guava in 2019 was 55 million tons. Guava fruit disease plays a significant role in both economic loss and guava production, as well as quality. Technology based methods allowing early identification of guava fruit disease symptoms. This is going to increase guava production and reduce financial losses. In this paper, we introduce an accurate guava fruit disease identification (GFDI) system using an improved convolutional neural network (Improved-CNN). This network based on eleven layers to extract features from the image dataset with Nesterov’s Accelerated Gradient (NAG) optimizer and Linearly Scored Categorical Cross-entropy (LSCCE) loss function. Additionally, preprocessing strategies such as crop function, data augmentation, noise removal, and splitting function have been applied to eliminate overfitting. Besides, we proposed a dataset of guava fruit diseases including phytophthora, root, and scab. According to the comprehensive experimental findings, the Improved-CNN model achieved accuracy ratings of 98% for trains and 93% for tests using a 0.001 learning rate. Additionally, the precision, recall, and f1-score have been determined as well. The proposed model performed significantly better than the state-of-the-art AlexNet model in the proposed dataset.