Enhanced Watermelon Leaf Disease Diagnosis through Attention – Fused DenseNet Architecture

Watermelon is a summer fruit which is one of the agricultural crops that have a lot of added value in worldwide. Early disease diagnosis for watermelon leaf disease is essential; depending on market demands and possible financial losses. In this paper, an Attention-fused DenseNet classification is proposed to quickly recognise watermelon leaf disease. Initially, using watermelon leaves disease dataset the denoising techniques reduces noise and maintain fine features using Adaptive Non local mean filter, to get high quality image. Next, the processed image is given to the morphological segmentation that transforms the leaf image based on their shapes by penetrating and modifying the image with a structural element (kernel). After that watermelon leaf image is featured by Scale Invariant Feature Transform (SIFT) to find the key points for further analysis. Finally, an Attention –fused DenseNet framework is processed to enhance the classification of watermelon leaf disease image. Using python software proposed framework have accuracy of 97% is accomplished when compared to other techniques.