Toward Efficient Image Recognition in Sensor-Based IoT: A Weight Initialization Optimizing Method for CNN Based on...

Area of Science:

• Computer Science
• Artificial Intelligence
• Image Processing

Background:

• Internet of Things (IoT) systems increasingly rely on visual data and deep learning for big data analysis.
• Traditional deep learning methods often overlook the significance of color information in image recognition.
• Developing efficient image data analysis methods for IoT remains a challenge.

Purpose of the Study:

• To propose a novel weight initialization method for deep learning image recognition that incorporates RGB color influence proportion.
• To enhance the training process of deep learning algorithms in visual data analysis for IoT applications.
• To provide an accessible method for determining early RGB influence proportions for practical use.

Main Methods:

• Extraction of RGB influence proportion from image data.
• Utilization of extracted RGB proportions in the weight initialization process for deep learning models.
• Experimental evaluation of the proposed method on various datasets.

Main Results:

• The proposed weight initialization method demonstrates effectiveness, particularly on small datasets.
• The method successfully integrates color information into the deep learning training process.
• An expedient approach for obtaining early RGB proportions was developed.

Conclusions:

• The RGB influence proportion-based weight initialization method offers a promising improvement for deep learning in image recognition.
• The method shows potential for application in IoT sensors for secure and complex data analysis.
• Further research could explore its scalability and application to larger datasets.