Product Defect Detection Based On Computer Vision

The quality of the product is related to the survival and development of the enterprise,and the quality of the drug affects people's health and even the stability of the society to a great extent.In the process of drug production,the quality defects of the bottle cap often exist in the vial as well as drug stratification,which affects the quality of the drug.In order to solve this problem,the paper gives a detailed analysis of the relevant technical of vials and drug quality detection,develops vision based algorithms for vials and drug quality detection,the algorithm can not only achieve real-time detection,but also can enhance the production speed,thereby increasing the income of the drug production enterprises.The main work of this paper is as follows: the proposed method of combination of threshold segmentation and image-blocking can obtain the region of interest of the image quickly and accurately;the characteristics of the sample are modeled and analyzed to generate the bag of words(BOF)model,make full use of the small intra class dispersion and big inter class dispersion of the feature points,obtain effective and low dimensional feature dictionary of the sample,namely eigenvector of the sample;the algorithm established two classification model of support vector machine(SVM)for vial cap and drug stratification defects under the condition of small samples,the defect detection of the unknown sample in this model can greatly improve the vials and drug identification accuracy and detection speed.The main research content is summarized as follows:1.The proposed method of combination of threshold segmentation and image-blocking can obtain the region of interest of the image quickly and accurately.If the whole sample is used as the research object of defect detection,the independent factors such as light and bottle body will affect the detection efficiency of the system.Because the reflection of different parts of the vial cap and the other parts of the bottle is different,the threshold segmentation algorithm is used to locate the defect area of the vial cap,and then the image is divided into 16?16 small blocks,and the small blocks with defects are the ROI of the image.The advantage of the algorithm is to improve the operation efficiency of the whole system,eliminate the influence of illumination and other parts of vial body on the overall detection efficiency and detection accuracy,and improve the robustness of the whole algorithm.2.The sample features are modeled and analyzed,and the BOF model of the sample is generated.In the BOF model,we put forward the algorithm of scale-invariant feature transform(SIFT),local binary patterns(LBP),histogram of oriented gradient(HOG),speeded-up robust features(SURF)to extract the feature vector of the ROI for each sample,and then combined with the k-means algorithm to find the clustering center of all samples,that is the visual vocabulary,and then generate visual dictionary of the samples,that is the final feature vector.The model makes full use of the characteristics of small dispersion in the sample feature point class and large dispersion between classes,and obtains a feature dictionary of low dimensional and effective samples.The experimental analysis shows that the algorithm can not only get the feature information quickly,but also maintain good robustness under the condition of large number of samples.3.Using machine learning algorithm,the algorithm established two classification model of support vector machine(SVM)for vial cap and drug stratification defects under the condition of small samples.Then the model is applied to defect detection of unknown samples.The experimental results show that the recognition accuracy of the model reaches maximum of 95%,and has a faster detection speed,which greatly improves the production efficiency of the enterprises.Support vector machine(SVM)has widely applicability and good portability,which breaks through the limitations of few samples and sample collection difficulties.