A Unified Efficient Deep Learning Architecture for Rapid Safety Objects Classification Using Normalized...

Area of Science:

• Computer Science
• Artificial Intelligence
• Industrial Safety Engineering

Background:

• Manual classification of personal protective equipment (PPE) in industrial settings is inefficient and time-consuming.
• Artificial intelligence (AI) offers a paradigm shift for object classification and tracking in complex environments.
• Existing methods struggle with the macro-level identification of personnel in intricate industrial spheres.

Purpose of the Study:

• To develop an efficient deep learning model for the rapid recognition and classification of PPE.
• To improve personnel safety in complex industrial settings through enhanced identification.
• To fuse the capabilities of multiple efficient deep learning models for superior feature learning and inference.

Main Methods:

• Exploration of several compact and efficient deep learning model architectures.
• Construction of a novel efficient model by fusing individual models based on contributory learning theory.
• Implementation of a normalized quantization-aware learning strategy for feature fusion.
• Development of a separable convolutional driven model as a base for combining architectures.

Main Results:

• The proposed fused model demonstrated rapid identification and classification of personnel and hardhats.
• Achieved remarkable speed and accuracy in classifying various hardhat classes in complex industrial settings.
• The normalized quantization-aware learning strategy effectively combined learned features from contributing models.

Conclusions:

• The developed deep learning model significantly enhances the efficiency and accuracy of PPE recognition.
• The fused model offers a practical solution for real-time safety monitoring in industrial environments.
• Normalized quantization-aware learning is a key contribution for creating accurate and rapid AI models.