Disease Detection in Potato Leaves Using an Efficient Deep Learning Model

• Published in: 2023 International Conference on Data Science and Network Security (ICDSNS) pp. 01 - 05
• Main Authors: Goyal, Bhanu, Kumar Pandey, Anil, Kumar, Rakesh, Gupta, Meenu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 28.07.2023
• Subjects: Analytical models; Convolutional Neural Network (CNN); Crops; Data models; Decision Support System(DSS); Deep learning; Deep Learning (DL); Network security; ResNet-9; Support Vector Machine (SVM); Support vector machines; Training
• DOI: 10.1109/ICDSNS58469.2023.10245369

Early blight and late blight are two diseases that put a massive hazard on potato crops and leave farmers feeling helpless. Early automatic recognition of these diseases will preserve time and enable farmers to act immediately on diseased crops. Machine learning and deep learning techniques offers variety of solutions for detecting rust in diseased crops. However, techniques of explaining such solutions are not common, but are essential because some machine learning models are considered to be as black boxes. In agriculture, the control of potato diseases is important because they lead to crop losses. Therefore, it should be identified and classified as a potato leaf disease to reduce losses. Doing it manually takes time. To solve the problems of mentioned above, an perfect and programmed technique for rapid identification and grouping of disease is desirable. In this study, machine learning and deep learning methods are used to organize potato leaves in two clusters using a present dataset called "Plant Village Dataset". To classify potato leaves into five groups, namely Potato Late Blight (PLB), Potato Early Blight (PEB) and Potato Healthy (PH) classes, this paper proposes a method based on algorithm of enhanced deep learning. The model is trained using the preexisting "The Plant Village" dataset, which consists of images with two diseases. To analyze the performance of the model, 1964 images of potato leaf disease have been collected from the Kaggle repository, out of which 54 sets of images are selected for training, 6 samples for data validation and 6 samples for testing, belonging to all categories. In the result analysis section, the SVM model performance is calculated at testing phase, and its correctness reached to 99.42%. Extensive testing has been made to demonstrate that our planned algorithm is more reliable and performs better than existing models in identifying and arranging potato leaf diseases.