Detection of whitefly pests in crops employing image enhancement and machine learning

Abstract

Agricultural research is currently undergoing a transformation with the emergence of precision agriculture, which utilizes automated monitoring, data collection, and analysis technologies. This new paradigm is expected to have a profound impact on agricultural practices, aiming to significantly improve both the quantity and quality of crop yields. One crucial challenge in precision agriculture is the automated detection of pests, as they can cause substantial damage to agricultural produce. However, the diverse nature of pests and the variety of crops they attack pose significant challenges for automated pest detection. A deep neural network-based approach has been proposed for the automated detection of whitefly pests in common plants. Before the actual training process, the captured images are subjected to contrast enhancement to ensure uniformity, as they are typically taken under varying lighting and partial shading conditions. The preprocessing step has been shown to enhance the accuracy of the proposed method by making the system more resilient to image degradations. The techniques utilized in this research employ decision tree (DT), convolutional neural networks (CNN), residual networks (ResNet), and attention-based CNN. The experimental results indicate that the proposed technique achieves accuracy rates of 81%, 96%, 97.5%, and 98% for the four models, namely DT, CNN, ResNet, and attention-based CNN, respectively. By comparing the results with those of baseline contemporary techniques, it is evident that the proposed model outperforms other deep learning models in terms of classification accuracy. Consequently, the method presented in this study can be considered an effective automated technique for accurately detecting whitefly pests and identifying pest infestations in crops.

A Deep Learning Methodology for Plant Species Recognition Using Morphology of Leaves

“…One of the significant impacts of deep learning lies in its effective feature extraction methods. Manual feature extraction processes require extensive expertise in the agricultural and ecological domains, and they are also time-consuming [11][12][13][14][15].…”