Compact, Optimized, and Effective: The CNNSpectra Approach to Deep Learning in Spectral Data Analysis

Deep learning has been instrumental in advancing various fields, including spectral data analysis. However, it is essential to recognize that deep layers and complex networks are not always prerequisites for achieving superior performance results. In this context, we introduce CNNSpectra, a simplified yet highly optimized deep learning model tailored for spectral data analysis, particularly for assessing food quality from portable spectroscopic devices. CNNSpectra’s performance is comprehensively evaluated on diverse datasets, including flour powders, spices, and chocolate, collected across multiple sessions. The results demonstrate that CNNSpectra consistently achieves competitive performance levels across various spectral datasets. While it may not outperform the best machine learning models in all cases, its simplicity and optimization render it a promising approach for real-world spectral data analysis, offering a valuable tool for food quality assessment from portable devices. This work underscores the importance of exploring more efficient and effective deep learning models in spectral data analysis and opens avenues for future research and improvements in this domain.