Research on Image Recognition Model of Tobacco Spare Parts Based on Multi-Feature Fusion

• Published in: 2025 10th International Conference on Intelligent Computing and Signal Processing (ICSP) pp. 550 - 556
• Main Authors: Zhang, Jing, Jin, Dongdong, Bian, Changzhi, Hu, Wei, Li, Zhongyin, Wang, Lichun, Fan, Jun, He, Linyang, Zhang, Bao
• Format: Conference Proceeding
• Language: English
• Published: IEEE 16.05.2025
• Subjects: Accuracy; Feature extraction; Feature fusion; Image classification; Image recognition; Printing; Production facilities; Signal processing; Text recognition; Tobacco spare parts; Training
• DOI: 10.1109/ICSP65755.2025.11087113

In order to realise the rapid query of tobacco spare parts, the tobacco spare parts recognition model of subject detection-feature extraction-feature retrieval was developed; the construction of multi-feature fusion of tobacco spare parts image recognition model (RMMF) realizes multi-feature fusion, extracts the image features through the spare parts image recognition model, distinguishes feature fusion, extracts the image features through the spare parts image recognition model, distinguishes different types of spare parts, and extracts the area features through the spare parts area retrieval model, and distinguishes different types of spare parts among the same type of spare parts. The RMMF was compared with three common image classification models, and the Top1 accuracy was 87.75% in the five-fold cross validation. compared with three common image classification models, and the Top1 accuracy of RMMF is 87.75%, the Top5 accuracy is 94.78%, and the Top10 accuracy is 96.74% in the five-fold cross-validation, which are better than the other models. RMMF also conducts two different types of case practice, for the 138 spare parts in the training set. RMMF's Top1 accuracy is 90.58%, Top5 accuracy is 97.10%, and Top10 accuracy is 97.83%; for the 5 spares that are not in the training set, RMMF does not misidentify them as spares in the training set, and the experimental results show that RMMF can achieve good results in real production.