Enhancing mango disease diagnosis through eco-informatics: A deep learning approach

Abstract

The mango is one of the most popular and economically important fruits in the world, but it is vulnerable to various diseases that can significantly reduce the quality and yield of the fruit. In the field of ecological informatics, tackling these challenges usually necessitate integrating cutting-edge technologies and ecological insights for effective disease management. Deep learning (DL) has been showing great promise in developing automated systems for leaf disease detection. This work proposes a novel DL approach that integrates the power of DL with ecological information management for the automated detection of mango leaf diseases. Our methodology introduces a visual modulation network that can innovatively learn the visual representations of leaf diseases along spatial and channel dimensions through simple but effective linear layers. An overlapped patching embedding is presented to tackle the discontinuity of non-overlapped patching in vision transformers (ViTs). Then, a succession of visual modulator blocks is presented to learn compact and long-term representations of disease in mango leaves without the inclusion of attention operation. Experimental results on the MangoLeafDB dataset show that the proposed visual modulation network can accurately detect various mango leaf diseases, demonstrating its potential as a cost-effective and efficient tool for disease management in mango cultivation.