This paper addresses contemporary food safety concerns by developing a non-destructive model that utilizes hyperspectral imaging (HSI) and artificial intelligence (AI) techniques. The goal is to improve the detection and classification accuracy of edible adulterants such as water, soda, glucose, and starch in milk. Initially, milk samples are prepared with three different concentration levels of edible adulterants for experimentation. A Resonon Pika L hyperspectral camera captures the milk data across 300 spectral bands, ranging from 400 to 1000 nm. In recent years, 3D deep learning models have attracted considerable attention for their capability to automatically extract nonlinear features from HSI datasets. This research demonstrates the effectiveness of the 3D-EfficientNet with Squeezed Excitation layer in detecting milk adulterants based on the spectral-spatial characteristics of the HSI data. The proposed model achieves an accuracy of 100% in detecting fresh milk across ten levels of adulteration, while the overall detection accuracy reaches 99.6%. Additionally, the study compares the results of existing machine-learning algorithms for milk adulteration detection based on reflectance values with the proposed model. Finally, the 3D EfficientNet-SE model with 99.6% accuracy outperforms the machine-learning approaches and other 3D deep models, offering a robust solution for detecting adulteration.