信息智能化技術(shù)的快速發(fā)展，對于貼片電容等電子元器件的需求量有了大幅度的提升，如何保證電容等電子類產(chǎn)品的安全性就需要對產(chǎn)品進行檢測，提高企業(yè)批量生產(chǎn)的質(zhì)量和效率。因此需要AI視覺通過無接觸、無損傷的實時檢測方法代替人工、傳統(tǒng)方式檢測，從而提升企業(yè)的生產(chǎn)率及產(chǎn)品質(zhì)量。

　　With the rapid development of information intelligent technology, the demand for electronic components such as chip capacitors has been greatly improved. How to ensure the safety of capacitors and other electronic products needs to be tested to improve the quality and efficiency of mass production of enterprises. Therefore, it is necessary for AI vision to replace manual and traditional detection methods by non-contact and non-invasive real-time detection method, so as to improve the productivity and product quality of enterprises.

　　像薄膜電容、貼片電容、電解電容等電子元器件生產(chǎn)過程中需要經(jīng)過復(fù)雜的工藝處理，在多重工序處理下，會出現(xiàn)各種問題，如表面缺陷、字符不清等。因為電子元器件種類繁多，各類電子元器件的結(jié)構(gòu)形狀、損壞程度和檢驗方法也均不相同，這就需要智能化的視覺檢測技術(shù)根據(jù)電容型號及特點進行定制化的檢測。

　　Such as film capacitors, chip capacitors, electrolytic capacitors and other electronic components in the production process need to go through a complex process, in the multi process processing, there will be a variety of problems, such as surface defects, unclear characters, etc. Because of the wide variety of electronic components, the structural shape, damage degree and inspection methods of various electronic components are different, which requires intelligent visual inspection technology to carry out customized detection according to the capacitor model and characteristics.

　　貼片電容元器件在生產(chǎn)過程中易出現(xiàn)孔洞、剝落、污點等缺陷，由于缺陷小，傳統(tǒng)算法需要耗費大量的時間對缺陷進行定制化開發(fā)，并且在進行灰度閾值分割時，易將微小的缺陷分割出去，很難保證在高速生產(chǎn)線上實現(xiàn)零缺陷檢測的要求；

　　Due to the small defects, the traditional algorithm needs a lot of time to develop customized defects, and when gray threshold segmentation is carried out, it is easy to separate the tiny defects, which is difficult to ensure the zero defect detection requirements in high-speed production line;

　　PCB板上存在很多焊點和細小零件，字符識別采集圖像時背景較為復(fù)雜，干擾因素多，造成字符定位和識別不準(zhǔn)確，加上零件本身反光，會出現(xiàn)識別信息不全、誤識別以及識別速度慢等情況，無法滿足實際生產(chǎn)檢測過程中對PCB板字符識別的需求；

　　There are many solder joints and small parts on the PCB board. The background of character recognition image acquisition is complex, and there are many interference factors, which lead to inaccurate character positioning and recognition. In addition, the part itself reflects light, resulting in incomplete recognition information, false recognition and slow recognition speed, which can not meet the requirements of PCB character recognition in the actual production and detection process;

　　PCB板在焊接元器件過程中，需要檢測每個元器件位置是否正確、元器件是否缺失等情況，傳統(tǒng)算法無法對多種電子元器件定位識別，而且定制開發(fā)需要耗費大量的時間，容易受外界因素影響，導(dǎo)致錯誤定位或元器件缺失，直接影響PCB板的性能及生命周期。

　　In the process of PCB welding components, it is necessary to detect whether the position of each component is correct and whether the components are missing. The traditional algorithm can not locate and identify various electronic components, and the customized development needs a lot of time, which is easy to be affected by external factors, resulting in wrong positioning or missing components, which directly affects the performance and life cycle of PCB board.

　　貼片電容元器件缺陷檢測系統(tǒng)只需在線上傳不同缺陷數(shù)據(jù)圖片進行標(biāo)注訓(xùn)練，即可準(zhǔn)確提取微米（μm）級的缺陷進行識別定位，從而實現(xiàn)高速流水線上零缺陷的目標(biāo)；通過專屬的神經(jīng)網(wǎng)絡(luò)架構(gòu)，通過準(zhǔn)確的標(biāo)注訓(xùn)練，完美適應(yīng)復(fù)雜背景下的字符識別，識別率高達99.99%。

　　The defect detection system of SMD capacitor components can accurately extract micron (μ m) defects for identification and positioning by uploading different defect data pictures online for annotation training, so as to achieve the goal of zero defect on high-speed pipeline; through the exclusive neural network architecture and accurate annotation training, it is perfectly adapted to character recognition under complex background, and the recognition rate is as high as 99.9 9%。

　　對不同的類型的貼片電容元器件進行定制化開發(fā)，只需上傳合格的產(chǎn)品圖片，對圖片內(nèi)的元器件進行訓(xùn)練學(xué)習(xí)，即可準(zhǔn)確定位PCB板上的元器件位置及完整性，在復(fù)雜的場景下?lián)碛懈玫男ЧR別速度可達毫秒級別。

　　For the customized development of different types of SMD capacitor components, we only need to upload qualified product pictures and train and learn the components in the pictures, then we can accurately locate the position and integrity of components on PCB board, and have better effect in complex scenes, and the recognition speed can reach millisecond level.