| New energy-based automatic pavement sweeping vehicles has potentially extensive application.To reduce energy consumption,the intelligent recognition system based on machine vision can improve the automation degree of the sweeper.This paper mainly studies the feature extraction and classification methods of garbage images,as well as the influence of contrast and noise on the performance of the algorithm.In the feature extraction stage,two technical schemes are proposed,namely traditional feature extraction and deep neural network based feature extraction.Traditional feature extraction operators include color,texture,gradient and frequency domain feature extraction,while deep neural network based feature extraction methods use pre-trained network as a feature extractor.In the classification method,the sparse representation and the kernel norm constraint optimization scheme are used.In the sparse representation method,the dual lagrangian multiplier method is used to solve optimization problem,and an improved iterative scheme is found;for the kernel norm constraint method,it uses the alternating direction multiplier method to reduce the iterative step size of the method by changing the objective function.Feature extraction and classification interact with each other,and both aspects must be optimized.According to different feature extraction methods and optimized iteration schemes,the accuracy and running time of the algorithm are evaluated,which shows that too many atoms are not conducive to the improvement on accuracy.During the evaluation of the adaptability of the algorithm,the accuracy of the algorithm under different image contrast and noise is analyzed.The results show that using deep neural networks to extract features not only exceeds the traditional feature extraction operator by 15% in accuracy,but also meets the real-time requirements.Meanwhile,the classification accuracy is less affected by image contrast and noise,and the data fluctuation range is within 5%. |
PDF Full Text Request