To address the challenges of complex components， difficult identification， and low intelligence of traditional detection methods for mixed explosives， two energetic material mixtures of m-dinitrobenzene/potassium nitrate and p-nitroaniline/ammonium nitrate were selected as research objects. A sequential detection strategy combining infrared preliminary screening and Raman confirmation was adopted. Combined with convolutional neural network （CNN）-based deep learning intelligent spectral image processing and recognition method， the spectral response characteristics of the samples in powder and flake forms were investigated. Meanwhile， the effects of component content and physical morphology on detection results were explored. The results indicate that for powdered energetic material samples， infrared spectroscopy can preliminarily identify the presence of m-dinitrobenzene and p-nitroaniline via characteristic peaks at specific wavenumbers， whereas it is difficult to independently distinguish inorganic oxidants such as potassium nitrate and ammonium nitrate. Raman spectroscopy can effectively characterize the nitrobenzene functional group structures of both powdered and flake samples. It can not only realize the qualitative identification of organic energetic components， but also detect characteristic signals unresponsive to infrared spectroscopy， thereby achieving accurate full-component identification of mixed explosives. Although instrumental parameters， excitation wavelength and sample morphology cause spectral peak shift and intensity fluctuation， the positions of core characteristic peaks and overall spectral profiles maintain favorable stability， which can provide a reliable spectral basis for the classification and identification of mixtures. The average recognition accuracy of the deep learning-based intelligent recognition model reaches 96.54% and 96.29% for mid-infrared and Raman spectral samples， respectively， with the average recognition time of a single sample being 0.044 s and 0.042 s.