The freshness of chilled mutton is influenced by various factors and can be comprehensively evaluated through multiple physical, chemical, and microbiological indicators. Traditional testing methods are complex and inefficient. Hyperspectral imaging technology, as a rapid and non-destructive detection technique, can effectively detect the changes in different components during the freshness variation of chilled mutton. To study the feasibility of using hyperspectral imaging technology for the multi-indicator evaluation of chilled mutton freshness, this paper proposes an improved artificial neural network (ANN) algorithm that enhances the correlation between labels by redefining the loss function and fully utilizes multiple freshness indicators to classify the freshness of chilled mutton. Experimental high-spectral images were collected for chilled mutton samples from 0 to 14 days in the 400 to 1 000 nm range. Laboratory methods were used to determine the values of total volatile basic nitrogen (TVB-N), pH value, total aerobic count (TAC), and an approximate number of coliforms (ANC) indicators. The original spectral data of chilled mutton samples were preprocessed using the S-G smoothing filter and multivariate scatter correction. The continuous projection algorithm (SPA) was used to select 18 feature bands of the spectral data as input data, and the proposed improved ANN algorithm was employed to establish a multi-indicator chilled mutton freshness grading model. The results showed that the improved ANN achieved a classification accuracy of 96% on the test set. The recognition rates for the three freshness levels of the samples were 100%, 89.28%, and 98.68%, respectively. The model was evaluated using four multi-label model evaluation metrics: Hamming loss, one-error, ranking loss, and coverage. The corresponding evaluation scores were 0.008, 0.002, 0.002 5, and 4.048, respectively. The accuracy and various model evaluation metrics of the improved ANN classification model were superior to those of traditional ANN, demonstrating the feasibility of using the improved ANN for non-destructive detection of multi-indicator chilled mutton freshness.