Image Defects Research Articles

AN APPLICATION OF A MATERIAL DEFECT DETECTION SYSTEM USING ARTIFICIAL INTELLIGENCE

In recent years, detecting defects in materials has become an important element in many industries, such as automotive, construction, textile manufacturing, and many others. Material defects, such as stains, cracks, dents, and others, can affect the quality and durability of the product, which can have serious consequences for safety, performance, and product quality. Therefore, it is important to detect and remove material defects as quickly and accurately as possible. Traditional methods of detecting material defects, such as manual inspections or simple visual algorithms, can be unreliable and, above all, time-consuming. In recent years, however, new tools based on intelligent algorithms (Liu et al., 2023) or structures (Neethu et al., 2015), such as neural networks, have emerged that enable automation and significantly increase the precision of defect detection. One of the most popular tools for object detection is the YOLOv5 (Redmon et al., 2016) (You Only Look Once version 5) model, which is based on neural networks and designed for fast and accurate object detection in images. The YOLOv5 model can be trained to detect specific types of material defects, such as stains, cracks, dents, etc. To use the YOLOv5 model for defect detection in materials, images of the material with specified objects of interest are needed, which will be used as input data for the model training. The YOLOv5 model analyses the images and determines rectangular frames around each defect indicating its location, and then assigns a selected label (in this case, the name of the defect). As a result, the YOLOv5 model is able to detect defects in new material images, regardless of their location, colour, and size. As the work progresses, it is also possible to prepare new images and retrain the model for new defects or to improve the performance of existing ones. As a result of the research, the obtained results at a 95% level of recognized defects certainly qualify the applied technology for professional use after appropriate workstation preparation and a sufficiently large amount of data (preferably 1000 or more).

Automated detection and segmentation of internal defects in reinforced concrete using deep learning on ultrasonic images

Periodic inspection of concrete structures for detecting internal defects through nondestructive imaging is recommended for ensuring structural safety and integrity. Ultrasonic imaging is one of the emerging and widely accepted techniques in nondestructive diagnostics of concrete structures. However, manually screening the ultrasonic images to detect and localize anomalies over large areas especially those from public infrastructure is tedious and prone to misjudgments. This work proposes a Region-based CNN to automatically detect, localize, and segment those signatures corresponding to defects from the noisy ultrasonic images with multiple features. This network replaces the conventional RoI pooling adopted in RCNNs with RoIAlign. To optimize the model for real-life complexity and noise levels, the network is trained on real experimental data. The ultrasonic image dataset has been generated by applying the imaging technique based on the Synthetic Aperture Focusing Technique on reflected waveform data obtained on various specimens with installed defects. The proposed network has been trained on the labelled image dataset containing signatures corresponding to multiple flaws like debonded rebars, delaminations, and planar crack-like defects in concrete members. The predictions of the trained model over the test datasets detect and masks the pixels corresponding to defects in ultrasonic images with a best mAP of 0.98. Segmentation results of the developed model upon testing are found superior to other state-of-the-art defect detection networks.

Application of infrared image defect detection based on data information visualization technology in intelligent physical education teaching

Application of infrared image defect detection based on data information visualization technology in intelligent physical education teaching

An improved meta heuristic IT2 fuzzy model for nondestructive failure evaluation of metal additive manufacturing lattice structure

Metal Additive Manufacturing (AM) lattice structure is widely used in various fields because of its lightweight and one-time forming characteristics. Nondestructive reliability evaluation still presents a challenge that requires immediate attention, especially in light of the service setting and when structural failure is used as an evaluation criterion. In this paper, the failure evaluation problem of structure is discussed based on the defects that affect yield stress, and the yield stress as the index. Firstly, a Computed Tomography (CT) scanner is used for non-destructive testing of metal AM lattice structure samples, and improved YOLO V7 is used to identify and count defects in CT scan images. The interval type 2 (IT2) fuzzy model is then enhanced using the irregular Gaussian function, and a meta-heuristic algorithm (particle swarm optimization) is added to enhance the model’s predictive capabilities. Finally, simulation and experiment are used to validate the AM lattice structure’s failure evaluation. The improved model’s superiority in the simulation exercise is demonstrated by contrasts with IT2 and the Type 1 fuzzy model. The yield stress prediction deviation of the sample during application verification is 1.12%, which attests to the validity of the failure prediction technique suggested in this paper.

АВТОМАТИЗАЦИЯ КОНТРОЛЯ КАЧЕСТВА НА ПРОИЗВОДСТВЕ С ИСПОЛЬЗОВАНИЕМ МЕТОДОВ МАШИННОГО ОБУЧЕНИЯ И КОМПЬЮТЕРНОГО ЗРЕНИЯ: РАЗРАБОТКА И АНАЛИЗ ЭФФЕКТИВНОСТИ СИСТЕМЫ

This article discusses methods for automating the quality control process in industrial enterprises using machine learning and computer vision. Special attention is paid to convolutional neural networks and their use for detecting defects in images. Specific examples of quality control system implementation in Python are given, describing the main steps in developing and using convolutional neural networks, as well as methods for evaluating system effectiveness. Overall, automating the quality control process using machine learning and computer vision is an effective way to improve product quality and optimize production processes.

Automatic detection of solar cell surface defects in electroluminescence images based on YOLOv8 algorithm

<p>Digital stethoscopes over recent years have gained acceptance among pulmonologists to perform auscultations due to their advantages over traditional stethoscopes. During the previous decade, researchers have prominently contributed to the development of algorithms aimed at enabling objective diagnosis of respiratory sounds and conditions, thereby affording individuals lacking medical expertise the capability to auscultate themselves. However, auscultation requires the per sonnel to be aware of the optimal chest position to place the device for a reliable diagnosis as well. This study aims to identify the optimal chest position to place a digital stethoscope’ s diaphragm to objectively diagnose chronic obstructive pulmonary d isease (COPD). Lung sound recordings from seven chest positions with data available in the ICBHI 2017 database namely, anterior left (Al), anterior right (Ar), lateral left (Ll), lateral right (Lr), posterior left (Pl), posterior right (Pr) and trachea (Tc), were analyzed in this study. COPD+ and COPD - at diagnosis, each chest position was done objectively using mel frequency cepstral coefficients (MFCC) features and machine learning models namely support vector machine (SVM) and decision tree (DT). The res ults indicate that the Pr chest position offers superior precision, recall, and F1 - score, with a recognition accuracy of 99.7% in COPD screening.</p>

On the Role of Image Processing Techniques in the Quantitative Image Analysis

Abstract The article explores how qualitative image analysis impacts the process of image interpretation, particularly in composite microstructure analysis. It highlights the importance of high-quality images for accurate computer-based object detection, emphasizing the limitations of rigid pixel-based rules compared to human visual perception. The study underscores the need for optimal imaging conditions to avoid image defects that hinder precise computational analyses in scientific and industrial applications.

LDD-Net: Lightweight printed circuit board defect detection network fusing multi-scale features

The current printed circuit board (PCB) defect detection model is difficult to balance accuracy and computational cost and cannot satisfy the requirements of practical applications. In this regard, a lightweight PCB image defect detection network (LDD-Net) with high accuracy is proposed in this study. First, LDD-Net designs a novel lightweight feature extraction network (LFEN), which adds convolutional layers for extracting large-scale feature maps and uses an efficient downsampling method to extract richer defect features. LFEN not only greatly reduces resource requirements, but also improves detection accuracy. Second, LDD-Net devises multi-scale aggregation network (MAN) to facilitate information sharing among feature maps at different scales. In addition, MAN incorporates an efficient attention module to suppress similar background interference. Finally, LDD-Net uses the lightweight decoupling head to detect three groups of shallow feature maps with rich information of the small object. Experimental results on the PCB defect dataset ,and VisDrone public dataset show that the mean average precision (mAP), recall, and precision of LDD-Net outperform state-of-the-art lightweight object detection models. The mAP, computation, number of parameters, and detection speed of LDD-Net are 95.90%, 21.48 G, 5.13 M, and 78.13 frames per second, respectively. LDD-Net effectively balances accuracy, computational cost, and detection speed for applications in edge systems or resource-constrained embedded devices.

A lightweight high-resolution algorithm based on deep learning for layer-wise defect detection in laser powder bed fusion

The quality of the powder bed is critical in the laser powder bed fusion (LPBF) process, and defects in the powder bed likely affect the quality of the final part. With the development of artificial intelligence technology, machine learning methods have been widely applied in powder-bed defect detection. However, to achieve high-precision defect detection, it is often necessary to construct complex network models and use high-resolution powder bed image data. To address these issues, this study used an off-axis industrial camera to capture layer-wise powder bed image data and proposed a defect detection model based on YOLOv7x and channel pruning to achieve defect identification and localization of powder bed patch images. Furthermore, an end-to-end defect detection pipeline based on image processing methods was proposed to detect defects in layer-wise powder bed images. Finally, the gradient-based class activation map technique (Grad CAM++) was used to analyze the interpretability of the detection results of the model. The results indicated that the proposed model was more lightweight than other models (YOLOv7x, Faster R-CNN, and SSD), with a model size of only 12.4MB. The average time for detecting powder bed image patches was significantly reduced to only 3.4 ms, and the average detection accuracy was as high as 97.4%. This demonstrates that the proposed detection method has the advantages of faster detection speed, higher detection accuracy, and simpler models, providing a reference for the real-time online detection of powder bed defects.

SEiPV-Net: An Efficient Deep Learning Framework for Autonomous Multi-Defect Segmentation in Electroluminescence Images of Solar Photovoltaic Modules

A robust and efficient segmentation framework is essential for accurately detecting and classifying various defects in electroluminescence images of solar PV modules. With the increasing global focus on renewable energy resources, solar PV energy systems are gaining significant attention. The inspection of PV modules throughout their manufacturing phase and lifespan requires an automatic and reliable framework to identify multiple micro-defects that are imperceptible to the human eye. This manuscript presents an encoder–decoder-based network architecture with the capability of autonomously segmenting 24 defects and features in electroluminescence images of solar photovoltaic modules. Certain micro-defects occupy a trivial number of image pixels, consequently leading to imbalanced classes. To address this matter, two types of class-weight assignment strategies are adopted, i.e., custom and equal class-weight assignments. The employment of custom class weights results in an increase in performance gains in comparison to equal class weights. Additionally, the proposed framework is evaluated by utilizing three different loss functions, i.e., the weighted cross-entropy, weighted squared Dice loss, and weighted Tanimoto loss. Moreover, a comparative analysis based on the model parameters is carried out with existing models to demonstrate the lightweight nature of the proposed framework. An ablation study is adopted in order to demonstrate the effectiveness of each individual block of the framework by carrying out seven different experiments in the study. Furthermore, SEiPV-Net is compared to three state-of-the-art techniques, namely DeepLabv3+, PSP-Net, and U-Net, in terms of several evaluation metrics, i.e., the mean intersection over union (IoU), F1 score, precision, recall, IoU, and Dice coefficient. The comparative and visual assessment using SOTA techniques demonstrates the superior performance of the proposed framework.

Low-field MRI at high magnetic field instability and inhomogeneity conditions

Understanding the effects of the magnetic field time instabilities in magnetic resonance imaging (MRI) is fundamental for the success of portable and low-cost MRI hardware based on electromagnets. In this work we propose a magnetic field model that considers the field instability in addition to the inhomogeneity. We have successfully validated the model on signals acquired with a commercial NMR instrument. It was used to simulate the image defects due to different types of instability for both the spin-echo and the gradient-echo sequences. We have considered both random field fluctuations, and an instability having a dominant harmonic component. Strategies are suggested to minimize the artifacts generated by these instabilities. Images were acquired using a home-made MRI relaxometer to show the consistency of the analysis.

Ensemble learning based defect detection of laser sintering

AbstractIn rapid development, Selective Laser Sintering (SLS) creates prototypes by processing industrial materials, for example, polymers. Such materials are usually in powder form and fused by a laser beam. The manufacturing quality depends on the interaction between a high‐energy laser beam and the powdered material. However, in‐homogeneous temperature distribution, unstable laser powder, and inconsistent powder densities can cause defects in the final product, for example, Powder Bed Defects. Such factors can lead to irregularities, for example, warping, distortion, and inadequate powder bed fusion. These irregularities may affect the profitable SLS production. Consequently, detecting powder bed defects requires automation. An ensemble learning‐based approach is proposed for detecting defects in SLS powder bed images from this perceptive. The proposed approach first pre‐processes the images to reduce the computational complexity. Then, the Convolutional Neural Network (CNN) based ensembled models (off‐the‐shelf CNN, bagged CNN, and boosted CNN) are implemented and compared. The ensemble learning CNN (bagged and boosted CNN) is good for powder bed detection. The evaluation results indicate that the performance of bagged CNN is significant. It also indicates that preprocessing of the images, mainly cropping to the region of interest, improves the performance of the proposed approach. The training and testing accuracy of the bagged CNN is 96.1% and 95.1%, respectively.

SGN-YOLO: Detecting Wood Defects with Improved YOLOv5 Based on Semi-Global Network

Object detection based on wood defects involves using bounding boxes to label defects in the surface image of the wood. This step is crucial before the transformation of wood products. Due to the small size and diverse shape of wood defects, most previous object detection models are unable to filter out critical features effectively. Consequently, they have faced challenges in generating adequate contextual information to detect defects accurately. In this paper, we proposed a YOLOv5 model based on a Semi-Global Network (SGN) to detect wood defects. Unlike previous models, firstly, a lightweight SGN is introduced in the backbone to model the global context, which can improve the accuracy and reduce the complexity of the network at the same time; the backbone is embedded with the Extended Efficient Layer Aggregation Network (E-ELAN), which continuously enhances the learning ability of the network; and finally, the Efficient Intersection and Merger (EIOU) loss is used to solve the problems of slow convergence speed and inaccurate regression results. Experimental results on public wood defect datasets demonstrated that our approach outperformed existing target detection models. The mAP value was 86.4%, a 3.1% improvement over the baseline network model, a 7.1% improvement over SSD, and a 13.6% improvement over Faster R-CNN. These results show the effectiveness of our proposed methodology.

AI-Generated Images for Speech Pathology-An Exploratory Application to Aphasia Assessment and Intervention Materials.

Images are a core component of aphasia assessment and intervention that require significant resources to produce or source. Text-to-image generation is an Artificial Intelligence (AI) technology that has recently made significant advances and could be a source of low-cost, highly customizable images. The aim of this study was to explore the potential of AI image generation for use in aphasia by examining its efficiency and cost during generation of typical images. Two hundred targets (80 nouns, 80 verbs, and 40 sentences) were selected at random from existing aphasia assessments and treatment software. A widely known image generator, DALL-E 2, was given text prompts for each target. The success rate, number of prompts required, and costs were summarized across target categories (noun/verb/sentence) and compared to frequency and imageability. Of 200 targets, 189 (94.5%) successfully conveyed the key concept. The process took a mean of 2.3 min per target at a cost of $0.31 in U.S. dollars each. However, there were aesthetic flaws in many successful images that could impact their utility. Noun images were generated with the highest efficiency and accuracy, followed by verbs, while sentences were more challenging, particularly those with unusual scenes. Patterns of flaws and errors in image generation are discussed. The ability to rapidly generate low-cost, high-quality images using AI is likely to be a major contribution to aphasia assessment and treatment going forward, particularly as advances in this technology continue.

Slice-Aided Defect Detection in Ultra High-Resolution Wind Turbine Blade Images

The processing of aerial images taken by drones is a challenging task due to their high resolution and the presence of small objects. The scale of the objects varies diversely depending on the position of the drone, which can result in loss of information or increased difficulty in detecting small objects. To address this issue, images are either randomly cropped or divided into small patches before training and inference. This paper proposes a defect detection framework that harnesses the advantages of slice-aided inference for small and medium-size damage on the surface of wind turbine blades. This framework enables the comparison of different slicing strategies, including a conventional patch division strategy and a more recent slice-aided hyper-inference, on several state-of-the-art deep neural network baselines for the detection of surface defects in wind turbine blade images. Our experiments provide extensive empirical results, highlighting the benefits of using the slice-aided strategy and the significant improvements made by these networks on an ultra high-resolution drone image dataset.

Automated Disease Severity Measurement Using Image Processing

Apple orchards are vital to the agricultural industry, but they face constant threats from diseases like apple scab, powdery mildew, rust, and rot. Traditional methods for measuring disease severity in apple leaves are labour-intensive, subjective, and prone to errors, limiting their practicality for large-scale agriculture. This paper introduces an innovative approach leveraging image processing techniques to automate disease severity measurement in apple leaves, offering a transformative solution for orchard management. The research aims to provide an efficient and accurate method for assessing disease severity, addressing the critical need for precise disease management and improved crop yields. Current methods relying on visual inspections by human experts are inefficient and unreliable. In contrast, the proposed method automates the measurement process, saving time and resources, enabling frequent monitoring, and enhancing early disease detection. This automation, in turn, leads to more effective disease management and increased crop yield and quality. The methodology involves image preprocessing, grayscale conversion, histogram equalization, thresholding, and defect area extraction, ultimately resulting in a quantitative disease severity index. The study utilizes a diverse dataset of apple leaf images, allowing for robust model development capable of accurate disease detection under varying conditions. This research holds significant implications for disease management and crop yield optimization, with broader applicability across agriculture. It paves the way for precision agriculture by enabling data-driven decisions on disease control measures while reducing the environmental impact of chemical inputs. Furthermore, the image processing techniques developed here can be adapted to other crops and plant diseases, promising efficient and accurate disease severity measurement methods across agricultural domains.

A Denoising Method for Multi-Noise on Steel Surface Detection

In order to solve the problem that steel surface defects are easily covered or submerged by other objects or noise, this paper proposes an open–closed transformation algorithm which can eliminate or weaken multiple noises. In the case of a small number of samples, this paper establishes a super-resolution generative adversarial neural network to achieve the enhancement of sample data. For avoiding unrealistic image defects caused by cuts or brightness variations, an enhancement method is given which incorporates the original defective high-frequency information into classical image fusion methods, such as rotation and error slicing. Experimental results show that the accuracy of the proposed denoising method reaches over 90%, which is more than 2.6% of that of the most primitive classification network. To compare with existing denoising methods, the denoising method proposed in this paper not only has higher accuracy, faster denoising speed, and stronger anti-interference ability, but also has better adaptation to the environment. This research will provide a new solution method for the denoising of multi-noise phenomena in multiple different environments.

Defect Classification for Additive Manufacturing with Machine Learning.

Additive manufacturing offers significant design freedom and the ability to selectively influence material properties. However, conventional processes like laser powder bed fusion for metals may result in internal defects, such as pores, which profoundly affect the mechanical characteristics of the components. The extent of this influence varies depending on the specific defect type, its size, and morphology. Furthermore, a single component may exhibit various defect types due to the manufacturing process. To investigate these occurrences with regard to other target variables, this study presents a random forest tree model capable of classifying defects in binary images derived from micrographs. Our approach demonstrates a classification accuracy of approximately 95% when distinguishing between keyhole and lack of fusion defects, as well as process pores. In contrast, unsupervised models yielded prediction accuracies below 60%. The model's accuracy in differentiating between lack of fusion and keyhole defects varies based on the manufacturing process's parameters, primarily due to the irregular shapes of keyhole defects. We provide the model alongside this paper, which can be utilized on a standard computer without the need for in situ monitoring systems during the additive manufacturing process.

Deep learning-based automated defect classification in Electroluminescence images of solar panels

The rise of deep learning approaches has boosted the modern computing capabilities and paved the way to novel outstanding applications. Recently, deep neural networks (DNN) have caught an increasing attraction in the field of PV cells detection and fault classification. Convolutional neural networks (CNN) were proved as DNNs that provide more reliable results in diagnosing these faults. On the other hand, Electroluminescence imaging (EL) analysis has been recognized as a powerful tool for addressing PV models malfunctions. In this work, an effective fault detection and classification approach is developed using multi-scale CNN-based models using two scenarios: a) a transfer learning-based approach using two selected DNNs (ResNet18 and ShuffleNet), and b) an independent light-depth CNN (denoted as CNN-ILD). CNN-ILD is a light CNN based on the insertion of parallel convolutional branches with low-to-high kernel sizes. This construction allowed CNN-ILD to capture a wide variety of features ranging from elementary to specific textures and patterns. The public ELPV dataset has been adopted after proper preprocessing and class categorization. The experimental results have shown promising classification results (88.41%-98.05%) of PV cell defects in EL images. CNN-ILD has a privilege in terms of the computational power/time and stability over the other two proposed pretrained DNNs.

MLR-Net: A multi-layer residual convolutional neural network for leather defect segmentation

It is critical to recognize anomalous features in wet-blue leather samples as part of industrial quality control in the leather industry. In industrial settings, manual inspection of leather samples is the present practice. Visual inspection-based leather processing is required to meet current industry requirements that encourage large-scale automation. Visual assessment of uneven surfaces is a difficult subject since irregularities can assume many different shapes and colors. The objective of this research is to develop an automated system that can detect the defects of leather images based on visual surface analysis. A deep learning-based strategy is developed to accomplish this goal, which trains the system to identify uneven and regular leather surfaces and classify leather images based on those surfaces. To address the task of defect segmentation in leather images, we introduce MLR-Net, a multi-layer residual convolutional neural network. Our proposed MLR-Net demonstrates promising performance in this task, achieving competitive results on our curated leather images dataset. The evaluation of MLR-Net reveals average sensitivity of 87.50%, specificity of 99.78%, an accuracy of 95.98%, F1-score of 85.90%, and Intersection over Union (IoU) of 78.98% respectively. These metrics indicate the effectiveness of MLR-Net in accurately identifying and segmenting defects in leather images.