Model Proposal for the Detection of Infected Potato Leaves Using Deep Learning

Abstract

In Peru, the cultivation of potatoes holds significant economic importance, serving as a cornerstone of the agricultural sector. However, the potato crop faces persistent threats from various diseases and pests, impacting both yield and quality. Addressing this issue is crucial for both the economy and food security in agricultural communities. Therefore, the study aimed to find the best classifier of infected potato leaves in the Peruvian context. For this purpose, the dataset titled’Data for: Identification of Plant Leaf Diseases Using a 9-layer Deep Convolutional Neural Network’ was chosen due to its relevance in the agricultural domain. This dataset comprises 39 different classes of plant leaf and background images, collected from various locations worldwide, totaling 61,486 images. Naturally, we applied preprocessing, which consisted of three phases: Preparation of Processing Tools, Pixel Values Normalization, Organization of Dataset. These phases were necessary to prepare the data and tools required for subsequent analysis, ensuring that the data are in an appropriate format and that the algorithms can work more efficiently and accurately. Finally, different Deep Learning models were implemented: InceptionV3, Resnet, and Vision Transformer. The results were evaluated according to the Accuracy, Precision, Recall and F1-score metrics. The best model resulted in Vision Transformer, whose metrics were superior to the others with 96.00% Accuracy, 96.18% Sensitivity, 96.00%, Precision and 95.94% F1-score. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.