Smart Detection of Crop Pests and Diseases: Enhancing Agricultural Productivity with AI and Modern Technology

Abstract

The world economy heavily depends on agriculture, especially in developing countries such as Pakistan, where it accounts for 19 per cent of GDP and provides 38 per cent of the labour force. Threats of plant diseases and pests in the sector are also critical, yet the percentage of losses to the harvest is close to 20 per cent of the total global harvest, with financial losses of over billions of dollars annually. Manual techniques for disease detection used in the past are labour-intensive, time-consuming, and, in most cases, are not available to rural farmers. The paper proposes a Deep Learning-based method for automated early detection and categorisation of plant health problems. Using such a large dataset of 55,449 image samples of healthy and diseased leaf datasets and certain pests such as the fall armyworm and leaf beetle, we designed a multi-model system which is combined with feature fusion. We use Convolutional Neural Networks (CNNs) as our method to process complex visual images and classify diseases across crops such as plantain, tomato, and cassava.The experiment's outcomes indicate some success. However, the model has obtained high F1-scores for each class (e.g., Class 12 with 0.72 and Class 19 with 0.59), the overall accuracy was 49% with a macro F1-score of 0.41. An elaborate analysis of the confusion matrix shows that there is a tough challenge in separating similar appearing diseases visually and managing the imbalanced classes of the minority datasets. The paper finds that, though deep learning has a disruptive potential in precision agriculture, future enhancements, such as smote to balance data, ResNet or Vision Transformer architectures, and multiple sensor fusion are necessary to cover the so-called lab-to-field gap and offer effective instruments to support sustainable farming.