This paper offers a prospective solution to the poor quality and less prominent features of the original terahertz spectral images of unsound wheat grains caused due to the imaging system and background noise. In this paper, a CBDNet-V terahertz spectral image enhancement model is proposed. Compared with the traditional algorithms, the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of the obtained enhanced images using the proposed model show performance improvement. As validated by the ResNet-50 classification network, the proposed model processes images with an accuracy of 94.8%, and the recognition accuracy is improved by 3.7% and 1.9%, respectively, compared to the images with only denoising and feature extraction. The experimental results indicate that the deep learning-based terahertz spectral image technology for unsound wheat kernels has good prospects in the identification of unsound wheat kernels.

In order to solve the problem of large image noise and unremarkable features caused by factors such as fluctuations in the power of a light source during the terahertz image acquisition of wheat, this paper proposes a THz image-denoising model called G-RRDB. Firstly, a module called Ghost-LKA is proposed by combining a large kernel convolutional attention mechanism module with a Ghost convolutional structure, which improves the characteristics of the network to acquire a global sensory field. Secondly, by integrating a spatial attention mechanism with channel attention, an attention module called DAB is proposed to enhance the network’s attention to important features. Thirdly, the Ghost-LKA module and DAB module are combined with the baseline model, thus proposing the dense residual denoising network G-RRDB. Compared with traditional denoising networks, both the PSNR and SSIM are improved. The prediction accuracy of G-RRDB is verified through the classification of the VGG16 network, achieving a rate of 92.8%, which represents an improvement of 1.7% and 0.2% compared to the denoised images obtained from the baseline model and the combined baseline model with the DAB module, respectively. The experimental results demonstrate that G-RRDB, a THz image-denoising model based on dense residual structure for moldy wheat, exhibits excellent denoising performance.

The traditional moldy wheat identification and detection method require complex processing steps, which take a long time and have less feature extraction ability, resulting in poor moldy wheat identification and detection. In this paper, a F-C-BLS terahertz spectral image recognition method for moldy wheat is proposed based on broad learning system. The F-C-BLS moldy wheat classification and recognition model is constructed to enhance the image quality and improve the network feature extraction. Experimental results show that the classification accuracy of our F-C-BLS network is 5.11%, 5.27%, 3.89 and 4.06% higher than that of BLS, RF, CNN and RNN, respectively. Therefore, our algorithm can effectively provide a new and effective method for the early identification of wheat mold.