Mold detection in cigarette tobacco is crucial for tobacco industry quality control. Traditional detection methods often suffer from limited efficiency, accuracy, and objectivity, highlighting the need for intelligent and reliable detection approaches. As a typical chemical sensing platform, the electronic nose (e-nose) system can generate characteristic responses to volatile organic compounds released during the mold development process of tobacco through a sensor array, providing chemical sensing information associated with mold contamination. However, single modal information is susceptible to environmental factors and sample differences under complex sample conditions, and its detection selectivity and stability still have limitations. In this study, we aim to develop an accurate and robust multimodal detection framework by integrating complementary odor and spectral information from an e-nose and hyperspectral imaging. Samples at three levels of mold contamination, namely, uncontaminated, moderately contaminated, and severely contaminated, were collected, and their e-nose responses and hyperspectral information were acquired. We designed a Complementary Feature Enhancement Module (CFEM) to strengthen single-modal feature representation and constructed a Cross-Modal Interaction Fusion (CMIF) module for deep inter-modal fusion. These components form the Collaborative Cross-Modal Interaction Fusion Network (CCIF-Net) for mold detection. Experimental results demonstrated that CCIF-Net significantly outperforms classical network models across all evaluation metrics while also validating the superiority of multimodal information over single-modal approaches. The framework provides an effective solution for industrial mold detection and offers valuable insights for multimodal fusion research in other domains.