Study On Prickly Ash Appearance Quality Detection Based On Machine Vision And SVM

Prickly ash is not only an important condiment, and also has a wide range of medicinal value. This paper focused on rapid detection of the appearance quality evaluation of prickly ash based on the machine vision technology and SVM, which included the detection of seed rate, fruit coat closing rate, rate of peduncle, and rate of seedcase and avoided using traditional manual method which encountered with some problems such as low efficiency and poor accuracy. The main contents and results of this paper were as follows:Firstly, building a hardware system for the detection of prickly ash's appearance quality. On the basis of experiments and analysis, choosing an image processor rationally, and constructing a hardware system, including computer, CCD camera, lens, lighting, motor, tray, light boxes, image acquisition card, etc.Secondly, carrying out research on the detection algorithm for prickly ash's appearance quality based on machine vision and SVM. The detection algorithm of prickly ash's appearance quality includes two parts, and image processing algorithm is foundation of modeling and predicting. In allusion to the early research of the laboratory, this paper optimizes and improves the detection algorithm of prickly ash's appearance quality, figures out a way for identifying single unit without second filling, and for the first time introduces the method of SVM into the predicting and classifying of prickly ash's inherent impurity.Thirdly, develping a software system for the detection of prickly ash's appearance quality on the basis of machine vision and SVM. Completing the software system for the detection of prickly ash's appearance quality on Visual C++6.0 platform, including image acquisition module, image processing module, feature recognition module, modeling and predicting module, and hardware controling module.Fourthly, analyzing the experiment of the detection of prickly ash's appearance quality. Random selecting 100 pepper samples from the 141 samples as sample set for modeling. Respectively collecting the image of seedcase, fruit coat closing, seed, and peduncle from each sample, and picking up the feature parameter as training set, and establishing prickly ash's classification model on the basis of LIBSVM software package. Selecting the remaining 41 samples as the prediction set. Acquiring the images and picking up individual feature parameter by processing the images. Then automatically forecasting the category of each unit on the basis of the established prickly ash's classification model, and counting the total number and the value of each detection index. Evaluating the value of each detection index of 41 samples of predicting set by the unaided eyes, and calculating the error and error rate of machine by comparing the two results.The results show that the identifying error rate is 9.09% for the seed,11.76% for the peduncle,50% and -4.04% for the fruit coat closing and the seedcase respectively.The results showed that this approach was efficient and credible, so it could help to establish the evaluation system of appearance quality detection of prickly ash theoretically and technologically.