RESUMEN

Anomaly detection has gained significant attention with the advancements in deep neural networks. Effective training requires both normal and anomalous data, but this often leads to a class imbalance, as anomalous data is scarce. Traditional augmentation methods struggle to maintain the correlation between anomalous patterns and their surroundings. To address this, we propose an adjacent augmentation technique that generates synthetic anomaly images, preserving object shapes while distorting contours to enhance correlation. Experimental results show that adjacent augmentation captures high-quality anomaly features, achieving superior AU-ROC and AU-PR scores compared to existing methods. Additionally, our technique produces synthetic normal images, aiding in learning detailed normal data features and reducing sensitivity to minor variations. Our framework considers all training images within a batch as positive pairs, pairing them with synthetic normal images as positive pairs and with synthetic anomaly images as negative pairs. This compensates for the lack of anomalous features and effectively distinguishes between normal and anomalous features, mitigating class imbalance. Using the ResNet50 network, our model achieved perfect AU-ROC and AU-PR scores of 100% in the bottle category of the MVTec-AD dataset. We are also investigating the relationship between anomalous pattern size and detection performance.

RESUMEN

In the field of automatic optical inspection (AOI), this study presents innovative strategies to enhance object detection accuracy while minimizing dependence on large annotated datasets. We initially developed a defect detection model using a dataset of 3579 images across 32 categories, created in collaboration with a major Taiwanese panel manufacturer. This model was evaluated using 12,000 ambiguously labeled images, with improvements achieved through data augmentation and annotation refinement. To address the challenges of limited labeled data, we proposed the Adaptive Fused Semi-Supervised Self-Learning (AFSL) method. This approach, designed for anchor-based object detection models, leverages a small set of labeled data alongside a larger pool of unlabeled data to enable continuous model optimization. Key components of AFSL include the Bounding Box Assigner, Adaptive Training Scheduler, and Data Allocator, which together facilitate dynamic threshold adjustments and balanced training, significantly enhancing the model's performance on AOI datasets. The AFSL method improved the mean average precision (mAP) from 43.5% to 57.1% on the COCO dataset and by 2.6% on the AOI dataset, demonstrating its effectiveness in achieving high levels of precision and efficiency in AOI with minimal labeled data.

RESUMEN

An integrated automatic optical inspection (iAOI) system with a procedure was proposed for a printed circuit board (PCB) production line, in which pattern distortions and performance deviations appear with process variations. The iAOI system was demonstrated in a module comprising a camera and lens, showing improved supportiveness for commercially available hardware. The iAOI procedure was realized in a serial workflow of image registration, threshold setting, image gradient, marker alignment, and geometric transformation; furthermore, five operations with numerous functions were prepared for image processing. In addition to the system and procedure, a graphical user interface (GUI) that displays sequential image operation results with analyzed characteristics was established for simplicity. To demonstrate its effectiveness, self-complementary Archimedean spiral antenna (SCASA) samples fabricated via standard PCB fabrication and intentional pattern distortions were demonstrated. The results indicated that, compared with other existing methods, the proposed iAOI system and procedure provide unified and standard operations with efficiency, which result in scientific and unambiguous judgments on pattern quality. Furthermore, we showed that when an appropriate artificial intelligence model is ready, the electromagnetic characteristic projection for SCASAs can be simply obtained through the GUI.

RESUMEN

In semiconductor manufacturing, defect inspection in non-patterned wafer production lines is essential to ensure high-quality integrated circuits. However, in actual production lines, achieving both high efficiency and high sensitivity at the same time is a significant challenge due to their mutual constraints. To achieve a reasonable trade-off between detection efficiency and sensitivity, this paper integrates the time delay integration (TDI) technology into dark-field microscopy. The TDI image sensor is utilized instead of a photomultiplier tube to realize multi-point simultaneous scanning. Experiments illustrate that the increase in the number of TDI stages and reduction in the column fixed pattern noise effectively improve the signal-to-noise ratio of particle defects without sacrificing the detecting efficiency.

RESUMEN

Automated optical inspection (AOI) plays a pivotal role in the quality control of contact lenses, safeguarding the safety and integrity of lenses intended for both medical and cosmetic applications. As the role of computer vision in defect detection expands, our study probes its effectiveness relative to traditional methods, particularly concerning subtle and irregular defects on the lens rim. In this research study, we propose a novel algorithm designed for the precise and automated detection of rim defects in contact lenses called "CLensRimVision". This algorithm integrates a series of procedures, including image preprocessing, circle detection for identifying lens rims, polar coordinate transformation, setting defect criteria and their subsequent detection, and, finally, visualization. The method based on these criteria can be adapted either to thickness-based or area-based approaches, suiting various characteristics of the contact lens. This approach achieves an exemplary performance with a 0.937 AP score. Our results offer a richer understanding of defect detection strategies, guiding manufacturers and researchers towards optimal techniques for ensuring quality in the contact lens domain.

RESUMEN

Training deep neural networks requires a large number of labeled samples, which are typically provided by crowdsourced workers or professionals at a high cost. To obtain qualified labels, samples need to be relabeled for inspection to control the quality of the labels, which further increases the cost. Active learning methods aim to select the most valuable samples for labeling to reduce labeling costs. We designed a practical active learning method that adaptively allocates labeling resources to the most valuable unlabeled samples and the most likely mislabeled labeled samples, thus significantly reducing the overall labeling cost. We prove that the probability of our proposed method labeling more than one sample from any redundant sample set in the same batch is less than 1/k, where k is the number of the k-fold experiment used in the method, thus significantly reducing the labeling resources wasted on redundant samples. Our proposed method achieves the best level of results on benchmark datasets, and it performs well in an industrial application of automatic optical inspection.

RESUMEN

This study proposes an adaptive image augmentation scheme using deep reinforcement learning (DRL) to improve the performance of a deep learning-based automated optical inspection system. The study addresses the challenge of inconsistency in the performance of single image augmentation methods. It introduces a DRL algorithm, DQN, to select the most suitable augmentation method for each image. The proposed approach extracts geometric and pixel indicators to form states, and uses DeepLab-v3+ model to verify the augmented images and generate rewards. Image augmentation methods are treated as actions, and the DQN algorithm selects the best methods based on the images and segmentation model. The study demonstrates that the proposed framework outperforms any single image augmentation method and achieves better segmentation performance than other semantic segmentation models. The framework has practical implications for developing more accurate and robust automated optical inspection systems, critical for ensuring product quality in various industries. Future research can explore the generalizability and scalability of the proposed framework to other domains and applications. The code for this application is uploaded at https://github.com/lynnkobe/Adaptive-Image-Augmentation.git.

RESUMEN

Given the huge demand for wire in today's society, the quality of the wire is especially required. To control the quality of the produced wire, the industry has a great desire for automated optical inspection technology. This technology is a high-speed and highly accurate optical image inspection system that uses mechanical sensing equipment to replace the human eye as the inspection method and simulates manual operation by means of a robotic arm. In this paper, a high-performance algorithm for the automated optical inspection of wire color sequence is proposed. This paper focuses on the design of a high-speed wire color sequence detection that can automatically adapt to different kinds of wires and recognition situations, such as a single wire with only one color, and one or two wires covered with aluminum foil. To be further able to successfully inspect even if the wire is short in the screen and the two wires are close to each other, we calculate the horizontal gradient of the wires by edge detection and morphological calculation and identify the types and color sequences of the wires in the screen by a series of discriminative mechanisms. Experimental results show that this method can achieve good accuracy while maintaining a good computation speed.

Asunto(s)

RESUMEN

Surface defects of fiber-reinforced resin matrix composites (FRRMCs) adversely affect their appearance and performance. To accurately and efficiently detect the three-dimensional (3D) surface defects of FRRMCs, a novel lightweight and two-stage semantic segmentation network, i.e., Mask-Point, is proposed. Stage 1 of Mask-Point is the multi-head 3D region proposal extractors (RPEs), generating several 3D regions of interest (ROIs). Stage 2 is the 3D aggregation stage composed of the shared classifier, shared filter, and non-maximum suppression (NMS). The two stages work together to detect the surface defects. To evaluate the performance of Mask-Point, a new 3D surface defects dataset of FRRMCs containing about 120 million points is produced. Training and test experiments show that the accuracy and the mean intersection of union (mIoU) increase as the number of different 3D RPEs increases in Stage 1, but the inference speed becomes slower when the number of different 3D RPEs increases. The best accuracy, mIoU, and inference speed of the Mask-Point model could reach 0.9997, 0.9402, and 320,000 points/s, respectively. Moreover, comparison experiments also show that Mask-Point offers relatively the best segmentation performance compared with several other typical 3D semantic segmentation networks. The mIoU of Mask-Point is about 30% ahead of the sub-optimal 3D semantic segmentation network PointNet. In addition, a distributed surface defects detection system based on Mask-Point is developed. The system is applied to scan real FRRMC products and detect their surface defects, and it achieves the relatively best detection performance in competition with skilled human workers. The above experiments demonstrate that the proposed Mask-Point could accurately and efficiently detect 3D surface defects of FRRMCs, and the Mask-Point also provides a new potential solution for the 3D surface defects detection of other similar materials.

RESUMEN

Plastic components play a significant role in conserving and saving energy. Plastic products provide some advantages over metal, including reducing part weight, manufacturing costs, and waste, and increasing corrosion resistance. Environmental sustainability is one of the sustainable development goals (SDGs). Currently, the non-contact computer-aided verification method is frequently employed in the plastic industry due to its high measurement efficiency compared with the conventional contact measuring method. In this study, we proposed an innovative, green three-dimensional (3D) optical inspection technology, which can perform precise 3D optical inspection without spraying anything on the component surface. We carried out the feasibility experiments using two plastic parts with complex geometric shapes under eight different proposed measurement strategies that can be adjusted according to the software interface. We studied and analyzed the differences in 3D optical inspection for building an empirical technical database. Our aim in this study is to propose a technical database for 3D optical measurements of an object without spraying anything to the component's surface. We found that the research results fulfilled the requirements of the SDGs. Our research results have industrial applicability and practical value because the dimensional average error of the two plastic parts has been controlled at approximately 3 µm and 4.7 µm.

RESUMEN

The plaque assay is a standard quantification system in virology for verifying infectious particles. One of the complex steps of plaque assay is the counting of the number of viral plaques in multiwell plates to study and evaluate viruses. Manual counting plaques are time-consuming and subjective. There is a need to reduce the workload in plaque counting and for a machine to read virus plaque assay; thus, herein, we developed a machine-learning (ML)-based automated quantification machine for viral plaque counting. The machine consists of two major systems: hardware for image acquisition and ML-based software for image viral plaque counting. The hardware is relatively simple to set up, affordable, portable, and automatically acquires a single image or multiple images from a multiwell plate for users. For a 96-well plate, the machine could capture and display all images in less than 1 min. The software is implemented by K-mean clustering using ML and unsupervised learning algorithms to help users and reduce the number of setup parameters for counting and is evaluated using 96-well plates of dengue virus. Bland-Altman analysis indicates that more than 95% of the measurement error is in the upper and lower boundaries [±2 standard deviation]. Also, gage repeatability and reproducibility analysis showed that the machine is capable of applications. Moreover, the average correct measurements by the machine are 85.8%. The ML-based automated quantification machine effectively quantifies the number of viral plaques.

RESUMEN

Rapidly and accurately detect the total nitrogen (TN) concentration is enormously important for surface water protection considering the critical role it plays in reflecting the eutrophication of surface water. However, traditional TN detection methods have to experience a tedious oxygen digestion process, which tremendously limits the detection speed of TN. To solve this problem, we propose a novel online rapid TN detection method. The transformations of nitrogenous substances during the oxidative digestion process are observed by using ultraviolet (UV) spectroscopy and the concentration of TN can be predicted by only using the variation of spectrum in the early oxygen digestion process. To select the most informative variables hidden in the collected three-dimension spectrum, a new wavelength selection algorithm called spatial interval permutation combination population analysis (siPCPA) is proposed, which considers the spatial-temporal relationships among each variable in the spectrum. By using the real surface water samples collected from Houhu Lake, Changsha, China, the effectiveness of our proposed new detection and selection methods are verified and compared with other state-of-the-art methods. As a result, the practical application experiment shows that our methods can determine the concentration of TN in 5 min with a relative error of less than 5%.

Asunto(s)

RESUMEN

In the traditional electroplating industry of Acrylonitrile Butadiene Styrene (ABS), quality control inspection of the product surface is usually performed with the naked eye. However, these defects on the surface of electroplated products are minor and easily ignored under reflective conditions. If the number of defectiveness and samples is too large, manual inspection will be challenging and time-consuming. We innovatively applied additive manufacturing (AM) to design and assemble an automatic optical inspection (AOI) system with the latest progress of artificial intelligence. The system can identify defects on the reflective surface of the plated product. Based on the deep learning framework from You Only Look Once (YOLO), we successfully started the neural network model on graphics processing unit (GPU) using the family of YOLO algorithms: from v2 to v5. Finally, our efforts showed an accuracy rate over an average of 70 percentage for detecting real-time video data in production lines. We also compare the classification performance among various YOLO algorithms. Our visual inspection efforts significantly reduce the labor cost of visual inspection in the electroplating industry and show its vision in smart manufacturing.

RESUMEN

Optical inspection is a rapid and non-destructive method for characterizing the properties of two-dimensional (2D) materials. With the aid of optical inspection, in situ and scalable monitoring of the properties of 2D materials can be implemented industrially to advance the development and progress of 2D material-based devices toward mass production. This review discusses the optical inspection techniques that are available to characterize various 2D materials, including graphene, transition metal dichalcogenides (TMDCs), hexagonal boron nitride (h-BN), group-III monochalcogenides, black phosphorus (BP), and group-IV monochalcogenides. First, the authors provide an introduction to these 2D materials and the processes commonly used for their fabrication. Then they review several of the important structural properties of 2D materials, and discuss how to characterize them using appropriate optical inspection tools. The authors also describe the challenges and opportunities faced when applying optical inspection to recently developed 2D materials, from mechanically exfoliated to wafer-scale-grown 2D materials. Most importantly, the authors summarize the techniques available for largely and precisely enhancing the optical signals from 2D materials. This comprehensive review of the current status and perspective of future trends for optical inspection of the structural properties of 2D materials will facilitate the development of next-generation 2D material-based devices.

RESUMEN

Infections of orchids by the Odontoglossum ringspot virus or Cymbidium mosaic virus cause orchid disfiguration and are a substantial source of economic loss for orchid farms. Although immunoassays can identify these infections, immunoassays are expensive, time consuming, and labor consuming and limited to sampling-based testing methods. This study proposes a noncontact inspection platform that uses a spectrometer and Android smartphone. When orchid leaves are illuminated with a handheld optical probe, the Android app based on the Internet of Things and artificial intelligence can display the measured florescence spectrum and determine the infection status within 3 s by using an algorithm hosted on a remote server. The algorithm was trained on optical data and the results of polymerase chain reaction assays. The testing accuracy of the algorithm was 89%. The area under the receiver operating characteristic curve was 91%; thus, the platform with the algorithm was accurate and convenient for infection screening in orchids.

Asunto(s)

RESUMEN

In this study, we developed a high-resolution, more accurate, non-destructive apparatus for refining the detection of electrode pixels in a thin-film-transistor liquid-crystal display (TFT-LCD). The hybrid optoelectronic apparatus simultaneously uses an array tester linked with the automatic optical inspection of panel defects. Unfortunately, due to a tiny air gap in the electro-optical inspector, the situation repeatedly causes numerous scratches and damages to the modulator; therefore, developing alternative equipment is necessary. Typically, in TFT-LCDs, there are open, short, and cross short electrical defects. The experiment utilized a multiple-line scan with the time delay integration (TDI) of a charge-coupled device (CCD) to capture a sharp image, even under low light, various speeds, or extreme conditions. In addition, we explored the experimental efficacy of detecting the electrode pixel of the samples and evaluated the effectiveness of a 7-inch opaque quartz mask. The results show that an array tester and AOI can detect a TFT-LCD electrode pixel sufficiently; therefore, we recommend adopting the hybrid apparatus in the TFT-LCD industry.

RESUMEN

Visual inspection is an important task in manufacturing industries in order to evaluate the completeness and quality of manufactured products. An autonomous robot-guided inspection system was recently developed based on an offline programming (OLP) and RGB-D model system. This system allows a non-expert automatic optical inspection (AOI) engineer to easily perform inspections using scanned data. However, if there is a positioning error due to displacement or rotation of the object, this system cannot be used on a production line. In this study, we developed an automated position correction module to locate an object's position and correct the robot's pose and position based on the detected error values in terms of displacement or rotation. The proposed module comprised an automatic hand-eye calibration and the PnP algorithm. The automatic hand-eye calibration was performed using a calibration board to reduce manual error. After calibration, the PnP algorithm calculates the object position error using artificial marker images and compensates for the error to a new object on the production line. The position correction module then automatically maps the defined AOI target positions onto a new object, unless the target position changes. We performed experiments that showed that the robot-guided inspection system with the position correction module effectively performed the desired task. This smart innovative system provides a novel advancement by automating the AOI process on a production line to increase productivity.

Asunto(s)

RESUMEN

This paper explores the effectiveness of the white, red, green, and blue light emitted diodes (LEDs) light sources to detect the third layer of the electrode pixel and the fourth layer of the via-hole passivation on thin-film transistors. The time-delay-integration charge-coupled device and a reflective spectrometer were implemented in this experiment. The optical conditions are the same, as each light source and the digital image's binary method also recognize the sharpness and contrast in the task. Consequently, the white and the blue LED light sources can be candidates for the light source for the optical inspection, especially for monochromic blue LED's outperformance among the light sources. The blue LED demonstrates the high spatial resolution and short wavelength's greater energy to trigger the photosensor. Additionally, the metal material has shown a tremendous responsibility in the photosensor with 150 Dn/nj/cm2 over the sensibility. The mercury 198Hg-pencil discharge lamp emits the stable spectral wavelength to significantly calibrate the spectrometer's measurement.

RESUMEN

This study aimed to develop an automated optical inspection (AOI) system that can rapidly and precisely measure the dimensions of microchannels embedded inside a transparent polymeric substrate, and can eventually be used on the production line of a factory. The AOI system is constructed based on Snell's law. The concept holds that, when light travels through two transparent media (air and the microfluidic chip transparent material), by capturing the parallel refracted light from a light source that went through the microchannel using a camera with a telecentric lens, the image can be analyzed using formulas derived from Snell's law to measure the dimensions of the microchannel cross-section. Through the NI LabVIEW 2018 SP1 programming interface, we programmed this system to automatically analyze the captured image and acquire all the needed data. The system then processes these data using custom-developed formulas to calculate the height and width measurements of the microchannel cross-sections and presents the results on the human-machine interface (HMI). In this study, a single and straight microchannel with a cross-sectional area of 300 µm × 300 µm and length of 44 mm was micromachined and sealed with another polymeric substrate by a solvent bonding method for experimentations. With this system, 45 cross-sectional areas along the straight microchannel were measured within 20 s, and experiment results showed that the average measured error was less than 2%.

RESUMEN

The paper presents a typology of electrical open and short defects on thin-film transistors (TFT) using an electrical tester and automatic optical inspection (AOI). The experiment takes the glass 8.5th generation to detect the electrical characteristics engaged with time delay and integration (TDI) charged-coupled-devices (CCDs), a fast line-scan, and a review CCD with five sets of magnification lenses for further inspection. An automatic data acquisition program (ADAP) controls the open/short (O/S) sensor, TDI-CCD, and motor device for machine vision and statistics of substrate defects simultaneously. Furthermore, the quartz mask installed on AOI verified its optical resolution; a TDI-CCD can grab an image of a moving object during transfers of the charge in synchronous scanning with the object that is significant.