Real-Time Target Detection Method Based on Lightweight Convolutional Neural Network.

Area of Science:

• Computer Vision
• Deep Learning
• Artificial Intelligence

Background:

• Deep learning advancements enhance target detection but often result in large models.
• Model size and detection speed are critical for practical applications in embedded systems.

Purpose of the Study:

• To propose a real-time target detection method using a lightweight convolutional neural network.
• To reduce model parameters and improve detection speed for embedded systems.

Main Methods:

• Developed a depthwise separable residual module combining depthwise separable convolution and non-bottleneck-free residual module.
• Replaced the VGG backbone in the SSD network with depthwise separable residual modules and convolutions.
• Integrated 1x3 and 3x1 convolution kernels to enhance feature extraction and create multiple detection feature maps.

Main Results:

• The proposed lightweight model significantly reduced parameter quantity and improved detection speed.
• Experimental results on complex scenes validated the method's effectiveness and superiority.
• Real-time performance was confirmed through video testing and Android platform deployment.

Conclusions:

• The lightweight convolutional neural network effectively addresses the challenge of large models in target detection.
• The method offers a scalable and efficient solution for real-time target detection in embedded systems.