Abstract: Fish is susceptible to temperature fluctuations during cold-chain transportation. Moreover, existing technologies lack sufficient capability for the early monitoring of quality deterioration, which easily leads to misjudgment of fish quality. In this study, a dual-colorimetric sensing tag combining a Time Temperature Indicator (TTI) and a pH-sensitive label was developed. Machine learning methods were employed to predict the freshness of unknown fish samples, and the optimized model was finally deployed on a WeChat Mini Program, enabling real-time scanning to obtain fish freshness results. Specifically, the TTI sensing unit mainly records temperature fluctuations during fish transportation and reflects freshness changes in the early storage stage; the pH sensing unit primarily monitors quality deterioration of fish in the middle and late storage periods. Furthermore, this study integrated the dual-colorimetric tag with machine learning. Model screening and construction were conducted based on monitoring data of salmon quality obtained by the tag, and the random forest algorithm was ultimately determined as the optimal model. In the validation test, the dual-colorimetric tag achieved an accuracy of 99.3%, representing an improvement of 5.23% and 6.71% compared with the single pH-sensitive label and single TTI label, respectively. To realize lightweight model deployment, an intelligent WeChat Mini Program for fish freshness detection was developed, which could rapidly output evaluation results of fish freshness from tag images. The testing accuracy reached 97.2% using 108 groups of actual image samples. The tag is fabricated from food-grade materials and can respond to multiple environmental factors synchronously, which enhances the stability and accuracy of freshness monitoring. It provides a novel strategy and method for the whole-process freshness monitoring of fish in cold-chain logistics.