A Lightweight Semantic Segmentation Algorithm Based on Deep Convolutional Neural Networks

Area of Science:

• Computer Vision
• Artificial Intelligence
• Deep Learning

Background:

• Convolutional Neural Networks (CNNs) offer high accuracy in image semantic segmentation but Deep CNNs (DCNNs) suffer from high memory consumption and latency.
• Real-time processing is crucial for applications like augmented reality and mobile interaction, posing challenges for existing DCNN models.

Purpose of the Study:

• To propose a novel semantic segmentation algorithm based on lightweight convolutional neural networks.
• To improve the speed of semantic segmentation while maintaining high accuracy.
• To address the limitations of existing DCNNs in terms of computational complexity and latency.

Main Methods:

• Developed a lightweight CNN model incorporating a position-attention mechanism to capture richer contextual information and model inter-pixel relationships.
• Introduced a channel attention mechanism in the decoding phase to enhance feature channel mining and improve low-level and high-level feature fusion.
• Evaluated the model's effectiveness on a publicly available dataset.

Main Results:

• The proposed lightweight CNN model achieved higher semantic segmentation accuracy compared to popular existing methods.
• The model demonstrated improved performance in objective evaluations, balancing segmentation accuracy with computational efficiency.
• The integration of attention mechanisms effectively enhanced feature extraction and fusion.

Conclusions:

• The proposed lightweight CNN algorithm offers a promising solution for efficient and accurate image semantic segmentation.
• The attention-based mechanisms are effective in improving contextual understanding and feature representation.
• This approach provides a viable alternative for real-time semantic segmentation in resource-constrained environments.