Multi-Crop Multi-Disease Classifier Using Hybrid Model

Rice and chilies are widely cultivated and consumed crops in Guntur district of Andhra Pradesh as well as India. However, recent statistics indicate that diseases such as bacterial leaf, blight rice blast, sheath blight, leaf spot, brown spot, leaf curl, whitefly, and yellowish leaf among others, significantly affect 10% to 15% of the yield, which impacts the national GDP by approximately 0.3315% of its 17%. Disease detection by using traditional methods, like visual inspection and laboratory testing is ineffective due to the requirements of domain expertise, work-force, and more time. To overcome these limitations, researchers concentrated on learning models, which can detect and classify diseases by analyzing spatial relationships and textures in images. Many researchers have focused on multi-disease classification in a single crop, whereas this work explored the multi-disease classification across multiple crops, like chili and paddy by transferring knowledge learned during the training of chilli using hybrid Inception-ResNet-V2, ResNet-50, and InceptionV3. The hybrid model is more suitable with a superior performance of 2-4% improvement compared to ResNet-50, Inceptionv3.