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

Abstract

Agricultural research is witnessing a new paradigm of technology employing automated monitoring, data collection and analysis termed as precision agriculture. It is expected that precision agriculture based applications would deeply impact the future methods of agriculture with an aim to substantially improve upon the existing magnitude as well as quality of the yields. One of the most sought after challenges in precision agriculture happens to be automated detection of pests which in certain cases can severely affect or ruin the agricultural produce. The divergences among the pests as well as the crops which they attack results renders serious challenge sin automated detection of pests. In this present work, deep neural network based approach for automated detection of whitefly pests in common plants, has been proposed. Prior to actual training, the images are contrast enhanced for sake of homogeneity among the images captured typically under varying lighting and partial shading conditions. It has been shown that the pre-processing increases the accuracy of the proposed work by making the system more robust to image degradations. The techniques employed in this paper employ the decision trees (DT), convolutional neural network (CNN), residual network (ResNet) and attention based CNN. The experimental results obtained show that the proposed technique achieves an accuracy of 81%, 96%, 97.5% and 98% for the four models respectively. Comparing the results of the proposed model with baseline contemporary techniques shows that the proposed model outperforms baseline deep learning models in terms of classification accuracy. Thus, the method proposed in the paper can serve as an effective automated technique for accurate detection of whitefly and pest infestation in crops.