Deep Learning-driven Precision Agriculture: Application of ResNet-20 for the Early Detection and Classification of Soybean Insect Pests under Real-world Constraints

Background: Soybean, a vital global oilseed crop, faces significant challenges from pest infestations and related damage that compromise productivity. Early and accurate detection of pest damage is critical for sustainable agriculture. This study evaluates the effectiveness of the ResNet-20 architecture in identifying and classifying soybean leaf conditions into three categories-Healthy, Caterpillar-damaged and Diabrotica speciosa-affected-and assesses its potential for practical agricultural deployment. Methods: The study utilized a dataset of 6,410 soybean leaf images sourced from the Mendeley database, categorized into three classes. The ResNet-20 model architecture incorporated convolutional layers, residual connections and global average pooling for efficient feature extraction and classification. Model performance was evaluated using metrics such as precision, recall, F1-score, accuracy and the Matthews Correlation Coefficient (MCC). Result: ResNet-20 demonstrated robust performance with an overall classification accuracy of 86.58%. High precision and recall values were observed for pest-affected categories, although a reduced recall for the “Healthy” class was noted. The MCC of 0.7728 highlighted strong predictive reliability. Misclassifications were attributed to visual similarities between pest-related symptoms and class imbalance in the dataset.