Apricot Plant Disease and Pest Detection from Field Images Using Fine-Tuned CNNs and Symptom–Organ Level Labeling

Early and accurate detection of plant diseases and pests is critical to preventing yield and quality losses, supporting sustainable agriculture, and ensuring food security. In this study, a novel dataset of 6,081 field images showing disease and pest symptoms on apricot (Prunus armeniaca) plants was created. Three pre-trained convolutional neural networks (CNNs), namely AlexNet, GoogLeNet, and ResNet-50, were fine-tuned for the classification task. Instead of a standard labeling strategy, a detailed labeling method was proposed, which considers both symptom type and the affected plant organ. The CNNs were trained on two datasets: a traditional 7-class version and a 13-class version generated using the proposed method. All models were evaluated using 5-fold cross-validation. Among all model and dataset combinations, the highest accuracy of 93.9% was achieved by the ResNet-50 model on the 7-class dataset. Although the proposed labeling method resulted in a slight decrease in classification accuracy, the performance difference remained small even with more classes. These findings indicate that the method is dependable and suitable for practical applications.