Multi-Modal Crime Scene Evidence Classification Using Transformer Networks

This research proposed the use of a transformer for classifying the multiple form of crime scene evidence including images, audio recording, and textual description. The model got an accuracy of 91 percent on the test set of the evaluation. 5%, a recall of 94.5%, and a recall of 87.8%, and an F1-score of 91.90%, and outperforms other machine learning architectures such as CNNs and RNNs. The results prove that transformers can learn complex features and combine several modalities making them a powerful tool to improve the accuracy of crime investigation process. The study also shows the effect of class imbalance problem and as well an advantage of data augmentation to enhance the performance of the model. Based on the results obtained above, we are inclined to propose that transformer networks can greatly contribute to the development of forensic analysis by offering more precise and detailed crime scene categorization.

Hali tarjima qilinmagan