Research on an industrial liquid level detection method based on an improved SSD.

Area of Science:

• Computer Vision
• Machine Learning
• Industrial Automation

Background:

• Industrial liquid level detection requires balancing computational efficiency and accuracy.
• Existing models often struggle with real-time processing on edge devices.
• Production incidents can arise from inadequate liquid level monitoring.

Purpose of the Study:

• To develop a lightweight Single Shot Detector (SSD) algorithm for edge devices.
• To enable real-time liquid level monitoring and anomaly early warning in industrial settings.
• To mitigate the risk of production incidents through improved detection.

Main Methods:

• A lightweight feature extraction network based on MobileNetV2 was developed.
• A Multi-Scale Dilated Inverted Residual Block (MS-DIRB) was proposed to reduce information loss.
• A Triplet Attention mechanism was integrated for enhanced feature learning.

Main Results:

• The proposed algorithm achieved 96.65% mAP, a 3.57% improvement over the baseline SSD.
• Model parameters were reduced by 81.58%, enhancing computational efficiency.
• The algorithm demonstrated suitability for real-time detection on industrial edge devices.

Conclusions:

• The lightweight SSD algorithm effectively balances efficiency and accuracy for industrial liquid level detection.
• The integration of MS-DIRB and Triplet Attention significantly enhances model performance.
• The proposed model is well-suited for real-time industrial edge computing applications.