Lightweight Neural Network for Traffic Sign Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traffic sign recognition algorithms based on convolutional neural networks require high computing power and storage, hindering their application on small mobile or embedded devices, necessitating lightweight treatments to reduce calculation costs and storage requirements.
Innovation Solution
A lightweight neural network method is developed, incorporating a convolution feature extraction part with separable asymmetric convolutions and a classifier part with separable full connection modules, followed by model pruning to reduce parameters and calculation amount while maintaining recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If convolutional neural network models are used for traffic sign recognition, then recognition accuracy is improved, but computing power and storage requirements increase
Solution Approach 1:
The neural network is divided into separate convolutional layers and fully connected layers, with each layer performing specific feature extraction and classification functions. This segmentation allows for optimized computation where convolutional layers handle spatial feature extraction efficiently while fully connected layers perform classification, reducing overall computational burden while maintaining accuracy.
Solution Approach 2:
The patent extracts and removes redundant parameters and unnecessary computational operations from the neural network model. Through parameter pruning and optimization, less important weights and biases are eliminated, reducing the model's computing power requirements while preserving the essential recognition capabilities needed for accurate traffic sign identification.
2Measurement precision
If convolutional neural network models are used for traffic sign recognition, then recognition accuracy is improved, but storage space requirements increase
Solution Approach 1:
The patent extracts and removes redundant parameters from the neural network model through pruning techniques. Unimportant weights and biases are eliminated, significantly reducing the model's storage space requirements while maintaining the essential features needed for accurate traffic sign recognition.
Solution Approach 2:
The patent optimizes the parameters of the neural network model by changing the scale, distribution, and structure of weights and biases. Through parameter normalization, weight sharing, and optimization, the model achieves reduced storage requirements without sacrificing recognition accuracy.
3Power
If lightweight treatment is performed on the neural network model, then calculation cost is reduced, but model accuracy may deteriorate
Solution Approach 1:
The patent applies parameter optimization techniques to the lightweight model, adjusting weights, biases, and hyperparameters to maintain high recognition accuracy despite reduced calculation cost. Through careful parameter tuning and optimization, the model achieves the best possible performance for its reduced computational capacity.
Solution Approach 2:
The patent implements feedback mechanisms during training and validation to monitor model performance and adjust parameters accordingly. Through feedback from validation data, the model continuously optimizes its parameters to maintain accuracy while operating with reduced computational resources.
Data Source
AI summary
Provided is a traffic sign recognition method based on a lightweight neural network, which including: a lightweight neural network model is constructed for training and pruning to obtain a lightweight neural network model; the lightweight neural network model comprises a convolution feature extraction part and a classifier part; the convolution feature extraction part includes one layer of conventional 3×3 convolution and 16 layers of separable asymmetric convolution. The classifier part includes three layers of separable full connection modules.