Deep Learning for Cotton Leaf Disease Detection

Cotton is among the most significant fibers grown in Indian subcontinent and it has an important role in the nation's economy. The cotton crops are susceptible to pathogens such as bacteria, viruses, and fungi. Most parts of the plant are infected, and yield losses are prevalent. The present research focuses on developing and optimizing CNN models to improve the cotton leaf disease detection and generate more accurate results. Transfer learning-based MobileNetV2 model has achieved high precision rates while keeping the parameters and calculation as low as could be expected. As the practical application of the model requires a time-efficient solution which ultimately gave better accuracy as well, the authors enhanced the MobileNetV2 model with a perception that the training time for the state-of-the-art MobileNetV2 could be reduced. In this research, a time-efficient optimized MobileNetV2 model is utilized to classify healthy and unhealthy cotton leaves and cotton plants. The performance of proposed model is compared with Inception V3, ResNet101V2, ResNet1S2, ResNet1S2V2, InceptionResNetV2, VGG16, VGG19, MobileNet, and MobileNetV2. The modified MobileNetV2 shows astounding accuracy of 99.91 which outperforms MobileNetV2 and other models. The proposed model has the potential to be compatible with mobile devices and would benefit the real-world farming.