Ecdyses serve as demarcation points between insect larval developmental stages, and the time required is a crucial reference for estimating the postmortem interval in forensic science, as well as for pest forecasting and control in agriculture and forestry. However, morphological and molecular methods typically used to estimate larval developmental duration are often inapplicable to the product of ecdysis, exuviae, making the development of new methods a significant challenge. This study provides, for the first time, developmental data on both the number and duration of larval instars in Dermestes coarctatus Harold, 1877, a forensically important necrophagous insect. Furthermore, 1770 spectral data acquired from 590 exuviae spanning nine instars were used to establish and compare six machine learning regression models: SVR, ANN, XGB-R, PLS-R, MLR, and CLR. The results indicated that as the instar of the exuviae increased, the absorption peak intensities of lipids and carbohydrates increased, while those of proteins decreased. This suggests that dermestid larvae enhance their chemical and mechanical barriers during development to adapt to environmental changes. The SVR model achieved satisfactory results with an R2CV of 0.91 and an R2P of 0.90, demonstrating high estimation accuracy. In summary, this study represents the first attempt to discriminate exuviae of different insect instars using attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy combined with chemometrics. It innovatively anchors the instar, developmental data, and spectral data of exuviae. This paradigm holds great promise for application not only to more necrophagous insects in forensics but also to other disciplines requiring insect developmental duration estimation.