Intelligent quantification of formaldehyde in aquatic product soaking solutions via a novel deep regression framework.

Introduction: The illicit use of formaldehyde (FA) as a preservative in aquatic products poses grave public health risks due to its proven carcinogenicity. While gold-standard techniques like HPLC offer high precision, their implementation in real-time surveillance is hindered by labor-intensive pre-treatment and reliance on expensive instrumentation, necessitating intelligent, non-destructive sensing platforms. Methods: We present Res-Reg, a novel image-to-value deep regression framework optimized for rapid FA quantification. The sensing paradigm integrates the high specificity of the Hantzsch chromogenic reaction with digital microscopic imaging, employing a residual-based backbone and a specialized regression head for multi-layer feature aggregation. To enhance environmental robustness against matrix interference, the training dataset was augmented with Gaussian blur kernels. Results: Systematic evaluation on 1,548 microscopic images demonstrated that Res-Reg achieved an MAE of 0.0027, MSE of 0.0028, and GM of 0.0002, significantly outperforming traditional CNN-based models by maintaining superior sensitivity in low-concentration intervals where optical signals are often obscured by turbidity. Discussion: Res-Reg effectively bridges the gap between microscopic optical signatures and precise chemical concentration without complex purification. This modular and scalable technology supports the digital transformation of food quality control, providing a potent solution for proactive safety surveillance and the preservation of global nutritional standards in the Food Industry 4.0 era.