An Effective Framework to Identify Diseases in Coffee Plants Leaves and Pesticide Recommendation System

Pesticides can be a valuable tool in the management of pests and diseases that infest crops. Pesticide advisory systems are crucial tools for modern agriculture, offering a number of benefits by ensuring accurate identification of pests and selective application, leading to reduced and efficient use of pesticides. Pesticide recommendation systems for coffee plants have limitations in accuracy that hinder effective pest management. To mitigate this limitation, this project presents a novel system which integrates Graph Neural Networks (GNNs) with ResNet-50. The input are images of coffee plant leaves, which are a binary input of "miner" and "rust" infested leaves. In the first step, a ResNet-50 model is trained on these images for 15 epochs with a classification accuracy of 98.93%. This ResNet-50 model is one that helps extract important feature vectors from the images. It, therefore, passes the extracted features into a GNN module, which is trained for 20 epochs before reaching a loss of 0.69418. The last stage includes the testing of the developed combined ResNet-GNN model, and consequent recommendation of the compatible pesticide treatment given the input image providing a relatively more accurate and effective solution for coffee plant pest control.