A novel framework for potato leaf disease detection using an efficient deep learning model

Abstract

Potato disease management plays a valuable role in the agriculture field as it might cause a significant loss in crops production. Therefore, timely recognition and classification of potato leaves diseases are necessary to minimize the loss, however, it is time taking task and requires human efforts. Thus, an accurate automated technique for timely detection and classification is needed to cope with the aforementioned challenges.There exist techniques grounded on machine learning and deep learning procedures that use the existing dataset i.e., ‘The Plant Village Dataset’ and perform classification into only two classes in potato leaves. Therefore, this article proposes a technique based on an improved deep learning algorithm that uses the potato leaf visual features to classify them into five classes i.e., Potato Late Blight (PLB), Potato Early Blight (PEB), Potato Leaf Roll (PLR), Potato Verticillium_wilt (PVw) and Potato Healthy (PH) class. The propose model is trained on the existing dataset i.e., “The Plant Village” that comprises of images having two ailments such as Early Blight (EB) and Late Blight (LB), and a Healthy class for potato leaves. Additionally, we have gathered the data for classes i.e., Potato Leaf Roll (PLR), Potato Verticillium_wilt (PVw) and Potato Healthy (PH) manually. A pre-trained Efficient DenseNet model has been employed utilizing an extra transition layer in DenseNet-201 to classify the potato leave diseases efficiently. Moreover, the usage of the reweighted cross-entropy loss function makes our proposed algorithm more robust as the training data is highly imbalanced. The dense connections with regularization power help to minimize the overfitting during the training of small training sets of potato leaves samples. The proposed algorithm is a novel and first technique to address and report the successful implementation for the detection and classification of four diseases in potato leaves. The algorithm’s performance was evaluated on the testing set and gave an accuracy of 97.2%. Various experiments have been performed to confirm that our proposed algorithm is more consistent and proficient to detect and classify potato leaves diseases than existing models.