Hybrid deep learning model for in-field pest detection on real-time field monitoring

The growth of important crops in agriculture can be affected and the production is reduced due to various pest attacks. The detection and recognition of these pests is a challenging task because of their identical look in the beginning level of plant growth. To overcome this challenge, deep learning-based real-time video detection models have been introduced for the segmentation and detection of different pests and pathogens. In this paper, a hybrid deep learning model is presented for the segmentation and detection of pests in various plants. The proposed technique is a four-stage model designed on the coordination of different deep learning networks. In the first stage, the image, as well as video frame, acquired images are denoised via the Bayesian image denoising framework. In the second stage, the denoised images are enhanced using LightenNet architecture. In the third stage, the image is semantically segmented with a context-guided residual network (ResNet) model. In the final stage, the segmented images are fed into the convolutional neural network to create a robust system for pest detection. The experiments are carried out on different benchmark datasets for performance assessment. The effectiveness of proposed method is verified in terms of structural similarity index measure (SSIM) and mean absolute error (MAE) and average precision (AP) as 0.99, < 0.2 and 89.67%, respectively. The qualitative performance evaluation of the proposed method indicates that it is apt for real-time monitoring and detection.