A Case Study on Attribute Recognition of Heated Metal Mark Image Using Deep Convolutional Neural Networks

Area of Science:

• Forensic Science
• Computer Vision
• Machine Learning

Background:

• Heated metal marks are crucial fire investigation evidence.
• Traditional analysis relies on physics and chemistry, posing challenges for qualitative assessment.
• Automated attribute recognition is needed to improve fire cause determination.

Purpose of the Study:

• To develop and evaluate a computer vision and machine learning model for recognizing attributes of heated metal marks.
• To establish a benchmark dataset for heated metal mark attribute recognition.
• To assess the model's performance across various parameters and conditions.

Main Methods:

• A benchmark dataset of heated metal mark images was created, selecting seven key attributes.
• A deep convolutional neural network (CNN) model was implemented for feature representation and classification.
• Extensive experiments were conducted, varying model structures, data augmentation, training modes, optimization, and batch sizes.

Main Results:

• The fine-tuned CNN model achieved high recognition rates for multiple attributes.
• Specific recognition rates included: metal type (0.925), heating mode (0.908), heating temperature (0.835), heating duration (0.917), cooling mode (0.928), placing duration (0.805), and relative humidity (0.92).
• Analysis included parameter influence, recognition efficiency, and execution time.

Conclusions:

• The proposed computer vision and machine learning approach effectively recognizes heated metal mark attributes.
• This method offers a preferable and potentially practical solution for fire investigation.
• The study demonstrates the utility of AI in enhancing forensic analysis of fire evidence.