Deep learning-based plant classification and crop disease classification by thermal camera
Ganbayar Batchuluun, Se Hyun Nam, Kang Ryoung Park
Abstract
Studies regarding image classification based on plant and crop disease images that were acquired using a visible light camera have been conducted in the past, whereas those based on thermal images are limited. This is because the thermal images are blurry due to the nature of the thermal camera, which makes it extremely difficult to classify objects. Therefore, this study proposes a new plant and crop disease classification method based on thermal images. The proposed method used a convolutional neural network with explainable artificial intelligence (XAI) to improve plant and crop disease classification performance. A new thermal plant image dataset was built for conducting the experiments, which contained 4,720 various images of flowers and leaves. In addition, an open database of crop diseases was also used, such as the Paddy crop dataset. The proposed plant and crop disease classification method demonstrated a 98.55% accuracy for the thermal plant image dataset and a 90.04% accuracy for the Paddy crop dataset, both of which outperformed other existing methods.