Improving Guava Disease Detection using Inception Mobilenet

Abstract

Guava, a tropical fruit highly valued for its nutritional benefits, contributes significantly to the economies of many developing countries, with the global guava market projected to reach ${\$}$ 8.8 billion by 2026. This fruit's economic importance is evident in its role in providing income to smallholder farmers and supporting local markets. However, guava production faces substantial challenges due to diseases such as anthracnose and wilt, which can reduce yields by up to 60 %. These diseases not only threaten the livelihoods of farmers but also disrupt the supply chain, affecting market stability and pricing. Recent advancements in technology are providing new solutions to these challenges. Deep Learning methods are having a significant impact on the detection and control of guava diseases, offering a promising approach to mitigate crop losses. A study proposed a novel model, InceptionMobilnet, which combines the strengths of InceptionV3 and MobilenetV2 architectures. After being trained on 527 images for 25 epochs, this model achieved an impressive accuracy of 99.90 % with recall, precision, and F 1-score of 0.9989, 0.9979, and 1, respectively, indicating the flawlessness of the model. Furthermore, Explainable AI (XAi) was employed to examine the regions of interest in the samples, enhancing the model’s transparency and reliability. Due to its low resource consumption nature, InceptionMobilnet can be converted into a mobile application, making it accessible to farmers and agricultural professionals in real-world settings. This innovative approach not only enhances disease detection but also empowers farmers with tools to protect their crops, thereby sustaining their livelihoods and contributing to the stability of the guava market.

Topics & Concepts