工業(yè)零部件智能視覺檢測(cè)設(shè)備
作為國內(nèi)外包裝智能自動(dòng)化設(shè)備研發(fā)企業(yè),上海陸甲自動(dòng)化科技有限公司的技術(shù)服務(wù)為中國制造業(yè)提供了與國際同步工業(yè)零部件智能視覺檢測(cè)設(shè)備技術(shù)解決方案。工業(yè)零部件智能視覺檢測(cè)設(shè)備應(yīng)用于:制藥、食品、飲料、日化、保健品、電子、電器、化工、汽車工業(yè)及塑料與五金等各大行業(yè)！
工業(yè)零部件智能視覺檢測(cè)設(shè)備在數(shù)字圖像處理技術(shù)是一個(gè)新興的技術(shù)行業(yè),已在自動(dòng)化系統(tǒng)、汽車零部件檢測(cè)和智能識(shí)別等領(lǐng)域都有的應(yīng)用。它已經(jīng)成為傳統(tǒng)人工檢測(cè)速度慢、檢測(cè)效率低的重要解決辦法之一。由于實(shí)際生產(chǎn)中,工業(yè)零件在細(xì)節(jié)方面會(huì)有諸多缺陷,因此,有必要選用合適的算法對(duì)其進(jìn)行準(zhǔn)確的識(shí)別和檢測(cè)。本文針對(duì)汽車吸能盒背板零件,設(shè)計(jì)了圖像檢測(cè)系統(tǒng)的整體方案,搭建了實(shí)驗(yàn)硬件平臺(tái),并詳細(xì)介紹了視覺系統(tǒng)采用的各種器件和照明系統(tǒng)的組成,再進(jìn)行攝像系統(tǒng)標(biāo)定,完成了畸變效應(yīng)的矯正。在獲取矯正后的圖像后,對(duì)圖像的預(yù)處理、邊緣檢測(cè)、零件幾何參數(shù)測(cè)量等關(guān)鍵技術(shù)進(jìn)行了重點(diǎn)研究。在預(yù)處理中,首先分析了圖像的噪聲類別,比較了多種濾波算法,找出適合本文圖像的濾波算法。進(jìn)而,在圖像邊緣檢測(cè)中,對(duì)比了經(jīng)典的邊緣檢測(cè)算法,為后續(xù)的特征提取提供了基礎(chǔ)。在檢測(cè)圖像基本特征時(shí),分別檢測(cè)圖像中的圓和直線,并對(duì)檢測(cè)結(jié)果的參數(shù)進(jìn)行了優(yōu)化,提高了圓和直線的檢測(cè)效果。在對(duì)圖像中的槽進(jìn)行檢測(cè)時(shí),采用了模板匹配算法,對(duì)槽的位置進(jìn)行了準(zhǔn)確的識(shí)別。在進(jìn)了了零件尺寸的檢測(cè)之后,文中還研究了完好零件、焊點(diǎn)零件和劃痕零件三種情況的分類識(shí)別方法。首先,通過邊緣檢測(cè),在保證圖像邊緣清晰、完整的基礎(chǔ)上,利用梯度方向直方圖算法進(jìn)行特征提取,并采用概率神經(jīng)網(wǎng)絡(luò)和SVM進(jìn)行分類識(shí)別,取得了不錯(cuò)的分類效果。然而,特征向量維度較高,特征提取信息混疊,以致圖像關(guān)鍵信息難以充分利用。文中對(duì)梯度方向直方圖算法進(jìn)行了改進(jìn),對(duì)梯度方向直方圖特征提取算法進(jìn)行雙線性插值,得到了更能夠體現(xiàn)細(xì)節(jié)特征的特征向量,再用神經(jīng)網(wǎng)絡(luò)和支持向量機(jī)進(jìn)行識(shí)別,在提高特征值抗混疊效應(yīng)的同時(shí),也提高了圖像的分類識(shí)別準(zhǔn)確率。本課題模塊的實(shí)現(xiàn)都是基于Visual C++和MATLAB的,包括視覺系統(tǒng)界面開發(fā)和算法的編寫。本文實(shí)現(xiàn)了零件特征的檢測(cè),與不同種類的零件分類識(shí)別。文中的研究結(jié)果體現(xiàn)了一定的工程價(jià)值,同時(shí)對(duì)圖像測(cè)量技術(shù)的應(yīng)用和零件的分類識(shí)別提供一定的借鑒意義。
Intelligent visual inspection equipment
As a well-known packaging intelligent automation equipment research and development enterprise at home and abroad, Shanghai Lujia Automation Technology Co., Ltd. provides technical solutions for the Chinese manufacturing industry to synchronize intelligent visual inspection equipment for industrial parts. Widely used in: pharmaceutical, food, beverage, daily chemical, health care products, electronics, electrical appliances, chemicals, automotive industry and plastics and hardware industries!
Intelligent visual inspection equipment for industrial components is an emerging technology industry in digital image processing technology. It has been widely used in automation systems, automotive parts inspection and intelligent identification. It has become one of the important solutions for slow manual detection and low detection efficiency. Due to the defects in the details of industrial parts in actual production, it is necessary to use an appropriate algorithm to accurately identify and detect them. In this paper, the overall scheme of the image detection system is designed for the back part of the car energy-absorbing box. The experimental hardware platform is built, and the components of the various components and lighting systems used in the vision system are introduced in detail. Then the camera system is calibrated and completed. Correction of distortion effects. After obtaining the corrected image, key technologies such as image preprocessing, edge detection and part geometric parameter measurement were studied. In the preprocessing, the noise class of the image is first analyzed, and various filtering algorithms are compared to find the filtering algorithm suitable for the image. Furthermore, in the image edge detection, the classic edge detection algorithm is compared, which provides the basis for the subsequent feature extraction. When detecting the basic features of the image, the circles and lines in the image are detected separately, and the parameters of the detection result are optimized to improve the detection effect of the circle and the line. When detecting the slot in the image, a template matching algorithm is used to accurately identify the position of the slot. After the inspection of the part size, the classification and identification methods of the intact parts, the solder joint parts and the scratch parts were also studied. Firstly, through the edge detection, on the basis of ensuring the image edge is clear and complete, the gradient direction histogram algorithm is used for feature extraction, and the probabilistic neural network and SVM are used for classification and recognition, and a good classification effect is obtained. However, the feature vector dimension is high, and the feature extraction information is aliased, so that the key information of the image is difficult to fully utilize. In this paper, the gradient direction histogram algorithm is improved, and the gradient direction histogram feature extraction algorithm is bilinearly interpolated. The feature vector which can reflect the detailed features is obtained, and then the neural network and support vector machine are used for recognition. The anti-aliasing effect of the value also improves the accuracy of classification and recognition of images. The implementation of all modules of this topic is based on Visual C++ and MATLAB, including visual system interface development and algorithm writing. This paper realizes the detection of part features and the classification and identification of different types of parts. The research results in this paper reflect a certain engineering value, and provide some reference for the application of image measurement technology and the classification and identification of parts.
