Res-DSCBAM A Comprehensive Framework for Efficient Oil Palm Pest Classification Using Integrated ResNet50, Depthwise Separable Convolution, and Convolutional Block Attention Module

Abstract

In this study, we provide a revised insight into the performance of the ResNet50 model in identifying pests affecting oil palm plants. This issue is particularly critical due to the significance of monitoring and early detection of pests to enhance oil palm productivity. This study aims to assess and enhance the performance of the ResNet50 model in identifying pests that affect oil palm plants. We aim to explore how the integration of Depthwise Separable Convolution and the Convolutional Block Attention Module (CBAM) techniques can enhance the model's accuracy and capability in effectively identifying pest classes. This study employed an experimental approach utilizing ResNet50 as the foundational model. The impact of incorporating Depthwise Separable Convolution and the CBAM was assessed to evaluate its effect on model performance. The experiments were conducted using a dataset that featured a diverse array of images depicting oil palm pests. The assessment of model performance involved a detailed examination of the Confusion Matrix and the classification report. The results surprisingly showed significant improvements in accuracy, precision, recall, and F1-score. The improvement was observed in each pest class, with the best final result achieved by combining both techniques, resulting in an average accuracy of 99.07%. This study demonstrated that the addition of Depthwise Separable Convolution and CBAM techniques significantly enhanced the ResNet50 model's ability to classify pests in oil palm. The results are noteworthy. However, additional analysis is required to identify the factors contributing to specific misclassifications. Future recommendations include exploring additional model architectures and further evaluating the factors that influence model decisions.