Terahertz (THz) spectroscopy and imaging systems have garnered significant attention in the field of material identification due to their non-destructive nature and ability to probe molecular structures. In this study, we employ THz time-domain spectroscopy (THz-TDS) to detect and image lactose content, aiming to distinguish its spectral signature in complex environments. While THz imaging provides high-resolution data, the presence of noise, especially in real-world scenarios, often hampers the accuracy of spectral analysis and image quality. To address this challenge, we propose an advanced image denoising approach using the Convolutional Blind Denoising Network (CBDNet), a deep learning-based neural network designed to mitigate noise in images with unknown noise distributions. The CBDNet model is trained to adaptively estimate the noise level and restore clear images from noisy THz data, improving the quality and interpretability of lactose images. Experimental results demonstrate the effectiveness of our method, significantly enhancing the clarity of THz images and the precision of lactose detection. This combination of THz spectroscopy with a state-of-the-art neural network presents a promising approach for improving image-based detection in biomedical and material analysis fields.