Classification of Plant Diseases using DenseNet-121 using Transfer Learning and Addition of Custom Layers for Efficient Performance

The prevalence of various diseases in plants caused by different disease-causing organisms such as bacteria, viruses, fungi, and many more, can drastically reduce crop yield, leading to concerns such as starvation. Reduced yield of crops due to plant diseases can exacerbate poverty due to farmers facing loss of income. Robust methods for the early detection of diseases in plants are crucial, as this can help to reduce the economic and environmental effects caused by the spread of diseases in plants. Diverse architectures of Convolutional Neural Networks have always shown their efficiency in image classification. In this study, diverse datasets have been integrated and a total of 6147 images of diseased plants have been considered for this work. This work demonstrates a novel approach to classifying plant diseases using a DenseNet-121-based CNN model by applying transfer learning and fine-tuning, along with custom layers addition on top of the pre-trained DenseNet-121 model architecture. Employing transfer learning techniques effectively reduces the training time of the Convolutional Neural Network architectures and improves accuracy through enhanced model generalization. The proposed model architecture has shown an accuracy of 0.978, thus underscoring the potential of automated systems in sustainable agricultural practices. Early detection of plant diseases can help prevent the decline in crop quality, thus contributing to enhancing food security. This research aims to address some of the Sustainable Development Goals.