Analysis Of Clutter Characteristics Based On Machine Vision

With the rapid development of optics,detectors and signal processing technology,the impact of noise on the performance of automatic target recognition(ATR)systems has become smaller and smaller,and background clutter has become the main factor affecting the performance of ATR systems.The purpose of the research based on machine vision clutter is to establish a scientific and effective clutter metric,quantitatively describe the difficulty of ATR system for target detection and reasonably evaluate the performance of ATR system.Firstly,this paper analyzes the basic theory of background clutter,including the basic concept and characteristics of clutter,the difference between human visual clutter and machine vision clutter,and the commonly used quantization metric of clutter.Then,the paper elaborates on a variety of image preprocessing algorithms,and selects the top hat filtering algorithm based on mathematical morphology,which has the best performance,through experiments,to preprocess the image,laying the foundation for the following research on the influence of image preprocessing algorithm on background clutter and target detection algorithm.In view of the fact that the current background clutter metrics are not suitable for establishing the relationship with the automatic target detection and recognition algorithm with machine vision as the core,this paper uses the two commonly used contours and textures based on the analysis of the ATR algorithm.The feature establishes the clutter metric based on the similarity of the target contour and the clutter metric based on the similarity of the gray level co-occurrence matrix.Both of these clutter metrics are global clutter metrics related to the target characteristics,and are more closely related to the performance of the ATR system.Using the two established clutter metrics and corresponding target detection algorithms for experiments,the quantitative relationship between background clutter and target detection performance(false alarm rate)was initially established,and the influence of top hat filtering algorithm on the clutter quantification and target detection algorithm is analyzed.The above clutter quantization and target detection are based on one feature of the target.Single features usually have limited access to target information,and multiple features can combine the advantages of single features and greatly improve the ATR system.Target detection performance.To measure the performance of the multi-feature ATR algorithm,it is necessary to establish a matching background clutter quantization method.Therefore,this paper finally uses the existing two single-feature clutter metrics and linear fusion ideas to establish a multi-feature clutter metric through the experimental comparison with the single feature clutter metric,the validity of the multifeature clutter metric is verified.