Novel Approach to Agricultural Pest Detection Leveraging an Enhanced ResNet50 Model for Insect Identification

Agriculture insect management is essential for crop production and food security since insects may cause major damage. Due to human error and laborious procedures, agricultural pest detection needs automation. The hybrid AI architecture in this work uses Convolutional Block Attention Module (CBAM) and Transformer layers in ResNet50 to improve detection accuracy in difficult field situations. The research contribution developed to fill the gap comes from the shortcomings of available models in handling large-scale, high-dimensional and light-sensitivity datasets of pests appearance due to facets illumination, environmental changes and pest transformations. This study introduces a hybrid artificial intelligence system with CBAM and Transformer layers on a ResNet50 backbone to improve insect identification in difficult agricultural situations. The research contribution developed to fill the gap comes from the shortcomings of available models in handling large-scale, high-dimensional and light-sensitivity datasets of pests appearance due to facets illumination, environmental changes and pest transformations. The model accuracies were 99.4% in distinguishing healthy from infected photos, which showed better performance than the VGG16 and DenseNet50 models with higher precision and recall towards most of the insect species tested. These results indicate that the proposed system can be used in operational agricultural pest monitoring systems providing an effective and scalable tool for pest management. The results show that hybrid deep learning may help the agriculture business manage sustainability.