Research On Defect Detection Of Infrared Capsule

The capsules are an indispensable part of our lives, their quality issatisfactory relationship to our health and safety, capsules defect detection isespecially important. The present system functions are detecting the shape of thecapsules and surface defects. Shape defects mainly is too long, too short ordeformation of the deflated head. Bubbles, folds, stains and other surface defects.This paper proposes a two-stage defect detection method of the capsule based oninfrared capsule image. The first phase extracting capsule shape and texture features,the BP neural network applied to the capsule qualified and unqualified recognitionclassifier. The second stage, using the fine detection method of the outline of theimage, identifying of the filtered capsules’ defects, such as bubbles, wrinkles, etc.This method was proven high detection accuracy, timely. The innovation of thispaper are summarized as follows:Firstly, in the system of image pre-processing stage, the segmentation resultsobtained diagram of basic global threshold method is connected between the capsuleand stuck part not well separated. In this paper, based on the smooth bimodalhistogram, proposes a method of selecting the threshold automatic in the split capsuletarget area and background stuck. After smoothing the histogram for imagesegmentation threshold automatic, select smoothed histogram from low to high, twotroughs position as a segmentation threshold. This method of detection segmentationeffect is superior to the traditional method, capsules regional and fixed capsules stuckwell separated, effectively removing the capsule stuck adhesion.Secondly, in this paper, the first position next detection method is used incapsules’ characteristic parameter extraction. While positioning of the capsule at bothends of the arc, positions the rectangle of the capsule both ends of the arc parts.When the capsule wall is detected, based on the basis of the conventional Hough linedetection, adding the weighting factor, which is the distance between the contour ofthe facts and the positioning standard contour. If the weighting factor is less than thepixel distance, then the point is considered to be the edge points, thus reducing the impact of the outlier detection result, the test proved that the method improves thedetection rate does not affect detection results under the premise.Thirdly, for this article detection capsule with different defect level, this paperproposes a two-stage detection method. The first stage, the choice of feature vectorfor the BP neural network, we will shape characteristics of the capsule and capsuletexture features combined and PCA dimensionality reduction as BP neural networkclassifier input, the preliminary identification capsules qualified or not. The secondstage, through qualified capsule edge detection, making an accurate judgment, thecapsule is determined qualified from the two aspects: the contour factor and thedistance from the capsules’ hat to body. The system process six capsules as a set,pretreatment each capsule take2499ms, recognition1000capsules consume0.1ms,rate of false positive1.6%, rate of recognition above95%.