Implementation and Performance Evaluation of Deep Learning Models for Disease Classification and Severity Estimation of Coffee Leaves

The cultivation area per farmer is consistently increasing while the labor force is reducing, due to which nowadays effective crop management is a major issue in agriculture. Effective crop management needs automatic techniques in farms. The crop gets damaged by biotic and abiotic stresses. Abiotic stresses include salinity, drought, heat, heavy metals, and cold, whereas biotic stresses include crop diseases and plant pests. These stresses affect plant growth and development and produce low-quality yields with reduced crop productivity. Biotic stress causes damage to plants by agents such as viruses, bacteria, fungi, and pests. Agricultural sustainability is a way of controlling biotic agents in an efficient manner and increasing productivity. Deep learning techniques allow correct and early identification of stress-causing agents, which helps a farmer to take preventive and corrective measures as early as possible to mitigate the problem. This paper presents a technique for training the deep neural network model to detect the disease symptom, classify the coffee leaves diseases, estimate the severity of identified diseases, and evaluate the performance of the developed system. The proposed model is a multi-task system for disease classification and severity estimation. In addition, we have experimented with new techniques of data augmentation for accurate results. Computational experiments for the proposed model are performed with several models such as DenseNet121, DenseNet169, DenseNet201, ResNet50, ResNet50v2, ResNet101, ResNet101v2, ResNet152, ResNet152v2, VGG16, VGG19, Mobile-Net, MobileNetv2, InceptionV3, InceptionResNetV2. Dense Net outperforms all other models. DenseNet169 obtained 92.421% classification accuracy and DenseNet121 achieved 62.302% accuracy for severity estimation. The experimental results specify the proposed model is an effective tool to assist farmers in the early identification of coffee leaf diseases.