Potato Leaf Disease Detection By Deep Learning: A Comparative Study

Potatoes are a well-known economic vegetable all over the world. It is the most economical food in a third-world country like Bangladesh. However, farmers must contend with several potato leaf diseases when cultivating this vegetable, including early and late blight. To detect potato leaf diseases, many convolutional neural network (CNN) approaches have been developed. However, these models have some limitations in identifying crop species or crop diseases generally since they were trained and validated on images of grower leaves from a particular area. In this study, a multiple transfer-learning model for identifying potato leaf disease was developed. This paper is a comparative study to detect potato images. The deep CNN model is employed in this study to detect disease from photos of potato leaves since CNN is utilized for image alignment and produces better results than others. This study employs nine transfer-learning deep CNN models: Inception V3, VGG16, VGG19, InceptionResNetV2, NasNetMobile, NasNetLarge, ResNet50V2, ResNet101V2, and DenseNet201 model. All of these models demonstrated greater than 90% accuracy, making them suitable for use in a variety of settings. DenseNet201 had the highest validation accuracy of 96% and train accuracy of 99% with low losses. This paper can help researchers and engineers to research potato frond disease detection. We used transfer-learning techniques to detect potato leaf disease with the highest accuracy and low loss among most researchers.