Research And Application Of Machine Vision Technology In Empty Bottle Inspection System

The research on the vision-based inspection system has been developed greatly and used widely in recent years, which can be definitely ascribed to its advantages, such as non-contact and real-time performance, high accuracy, and so on. The empty bottle inspection system is the vision based intelligent inspection equipments integrated of optical, mechanical and electronical functions. The system performs well in quality inspection of empty bottle before beer filling instead of manual work. As the food safety draws more and more attention in China, and the foreign equipment for empty bottle inspection can't meet the quality situation of the recycled bottles in our country, the research on empty bottle inspection system with full independent intellectual property rights becomes critical important, which willmeet the needs of the beer industry and the market.In this paper, the location of empty bottle inspection system on liquid filling lineis described. The design of the high-speed online bottle inspection system based on machine vision technology, the system architecture and principles are provided in details. Then some deep research on the machine vision system and key technologies for the image recognition system are presented, as well as an intelligent inspection algorithm for the recognition of the finish, base and sidewall of empty bottle. The contributions are summarized as follows:1. This paper analyzes the optical characteristics of the bottle flaws and studies the key technologies of image acquisition. Then it designs an image acquisition system providing the clear images for finish, base and sidewall of empty bottles respectively. Particularly the paper presents an optical imaging setting to implement omnidirectional image acquisition of sidewall. For the purpose of solving the problem that heavy-tailed images were captured as bottles moving under high-speed, the paper proposed a method with stroboscope cooperating with shutter speed.2. The main constraint is the real time operation as the bottles move continuously on the conveyer belt. Using images obtained from the camera, we just focus on the image area reflecting the characteristic of bottles. Then the proposed algorithm obtains the position of the bottles in the image and tracks the detection area precisely. The method is proved to be efficient in improving inspection speed.3. Based on the analysis of the characteristics of the finish bottle images, which shows the circle profile and the complexity of finish images, an algorithm derived from least-square circle fitting is applied for real-time track of finish areas in the image. For flaw feature extraction moment feature of connected components is adopted. After that a proposed algorithm based on feature statistical decision analysis is used for quality classification. It can deal with combination noise problem and obtain the satisfying accuracy with lower computational load. The proposed algorithms are proved to be efficient in automatic finish inspection.4. Base and sidewall cracks are regarded as critical defects for the recycled bottles adopted in beverage production. The sidewall axis is determined by scanning points of maximum grayscale grads on the line. And detection area of sidewall can be tracked for real time with high accuracy with the method of using symmetry analysis and probability statistics. The improved least-square fitting is used in base tracking. The bottles for any defects or presence of alien objects are checked using flaw feature extraction and statistical decision, and the ones with any defect are rejected and separated from the main stream. The proposed algorithms are proved to be efficient in the automatic base and sidewall inspection.5. The key technologies for the process of the empty bottle inspection system with the advantages of high-accuracy, integration, real-time and high-speed characteristics are summarized, which includes the proper configuration of hardware and software, the quick and efficient algorithm and the independent parallel performance of models.