Identifying Plant Diseases with Image Recognition

Abstract

Worldwide, approximately 70-80% of plants suffer from some sort of plant disease [1]. These diseases ravage through billions of dollars worth of crops and thousands of tons of food, causing devastation to local economies and increasing food insecurity throughout the globe. However, the use of AI technology can help combat plant diseases by early recognition of diseases through image recognition. This study employs a convolutional neural network (CNN), implemented in Python, which is trained on the “New Plant Disease Dataset,” published on Kaggle, to classify different plant diseases [2]. We used a subset of this dataset across three folders. The training folder had ~16,000 images, the validation folder had 3813 images and the testing folder had 1673 images. There were 8 unique labels on which the model was trained on. The images are of healthy and infected leaves of these crops. The trained model achieved an accuracy of 89.73% in testing but achieved 96.73% accuracy when tested against the validation folder in the training process. Importantly, the classification was not a binary prediction of healthy versus infected plants, but classified the specific crop and specific type of disease. Interestingly, most of the misclassification was between healthy versions of different crops, and the model was even more powerful when considering just its ability to predict diseases. This study highlights the potential use of CNNs in automated disease detection. Thus, the use of AI methods can contribute towards mitigation of agricultural losses and enhanced food security.