Prediction and classification of wheat plant disease in Pakistan using deep learning techniques

Abstract

Agriculture is the backbone of the economy in many developing countries, including Pakistan. The most economically important crops in Pakistan are wheat and cotton. However, the production of these crops is severely affected by diseases caused by pests, pathogenic bacteria, and viruses, which significantly impact a country's economic growth. Over the last decade or so, numerous image-processing techniques and deep-learning models have been developed by researchers for the identification of plant diseases; however, wheat plant disease has been very rarely addressed. In this research, several deep learning models were trained on a dataset of wheat plant disease images, which were collected from research labs and by taking images through Android mobile cameras from different regions of Khyber Pakhtunkhwa and Punjab. The dataset contains 650 images with 5 classes of wheat diseases, which is increased to 3500 images by using the classical augmentation technique to resolve the problem of overfitting. Among all the deep learning models, ResNet50 and VGG16, with fine-tuning, achieve outstanding accuracies of 95% and 93% on test data, respectively, which outperform the results of previous research conducted by the researcher using WDD17 and other datasets.