Design And Implementation Based On Color Feature Extraction Of Pepper Automatic Classification System

With the development of science and technology and the progress of computerhardware, computer vision technology has been extended to many aspects of the aerospace,military, medicine, and people’s daily life, and some studies have been done on recognition,inspection, classification and measurement of the crops by computer vision technology.However, growth of the crops is closely affected by different conditions like growingenvironment, soil, moisture, light, etc, resulting in each test sample having similarcharacteristics and individual similarities and differences, therefore, recognition of the cropsface a higher demand than which of industrial parts. Considering the above characteristics ofcrops, in the field of agricultural visual recognition, it is necessary to put forward a moreuniversal, targeted graphics image processing algorithms in order to extract thecharacteristics of target samples in the complex conditions.China is a big country of pepper production and export and the pepper production andprocessing requirements continue to expand. It is worth to point out that pepper classificationand acquisitions are essential links in the production and processing and the pepper qualityand grade directly affect the pepper economic benefits. Currently, the pepper qualityclassification in the acquisition process still mainly relys on manual work which has manydrawbacks such as low classification efficiency, subjective factors, human relations, etc.Therefore, in order to meet the needs of the pepper automatic detection and qualityclassification, we studied the automatic classification which includes the analysis pepperexternal characteristics based on color feature, and built a hardware system of pepperautomatic classification in quality testing, realizing the automatic detection and classificationof pepper upon multiple indicators like color characteristics, external defects, lesions, etc.The contents of this paper are as follows:1. We studied pepper quality testing system based on computer visiontechnology. In the system, we provide the pepper image samples unified light intensity andbackground color, In this way, we can exclude some interferences of external factors likelight, background to further analyse the the contour and color characteristics of samples.2. The method of image processing was used to preprocess the collected pepper images,mainly including foreground and background separation, image enhancement, binarymethod, multiple erosion and dilation, the K-means clustering analysis. The Characteristic of peppers’ smooth surface can cause high-light-areas in imaging process.These high-light-areas, which are generally small circular bright spots with yellow color,producing shaded areas in flute groove position, are highly similar with pepper lesion images.Thus, the production of high-light-areas will cause serious impacts on pepper lesionidentification using machine vision and the production of shaded areas will bring somedifficulties in peppers’ foreground extraction. This paper proposes two solutions;(1) usingmulti-points light source which will produce the diffuse reflected light to irradiate samples toreduce the production of high light and shaded areas in the natural light irradiation.(2) Usingmethod of K-means mean cluster analysis to completely eliminate the disturbance ofhigh-light areas. In these ways, the proportion of pepper lesions can be automaticallyanalyzed and counted and more accurate analysis of pepper classification can be obtained.3. The main factor that impacts pepper classification is the proportion of externalcharacteristics which are associated with the pepper colors, thus, it is obliged to designexperiments to find out the quantitative relationship between pepper classification andpepper color, so that pepper classification could be identified through pepper color featureextraction. In our work, we use color histogram method to extract pepper color features.4. Vector machine method was used to study known pepper pictures of six grades with50pictures in each group. Then, we established optimal classification hyperplane using thepercentage of lesions, flower leathers and yellow shoot in major pepper color features as theindex to formation classification model library.5. Treated through grade sample library screening examination pepper set thresholdlevels, the use of machine vision technology for unknown levels of180chilli automaticimage classification, the classification results and the results compared artificial grading,inspection machines for bulk pepper instead of doing experimental effect automatic grading.