Defect Detection Method Based on Knowledge Distillation

Zhou, Qunying; Wang, Hongyuan; YingTang,; Wang, Yang

doi:10.1109/access.2023.3252910

Cited by 12 publications

(5 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, Sizhe Xiao et al [13] proposed a gradient-based unsupervised model, Grad MobileNet, based on MobileNetV3, in which the model the model can be trained using only a few normal images, extracting the feature gradient of the input image, classifying welding defects through the gradient distribution, and achieving 99% accuracy on the welding defect dataset RIAM, which was constructed by the authors. Qunying Zhou et al [14] proposed a knowledge-distillation model based on attention mechanism and feature fusion, which enhances the ability of the model to extract features through attention, improves the pixel-level localization of the model, and provides better detection results in the MVTecAD dataset. Jin Rui et al [15] proposed a fabric-defect-detection method based on an improved generative adversarial network, introducing a center loss constraint to improve the recognition performance of the method, which was evaluated on the publicly available Tianchi dataset with good results.…”

Section: Related Work 21 Deep Learning Detection Methodsmentioning

confidence: 99%

A Feature-Oriented Reconstruction Method for Surface-Defect Detection on Aluminum Profiles

Tang,

Zhang,

Jin

et al. 2023

Applied Sciences

View full text Add to dashboard Cite

The number of defect samples on the surface of aluminum profiles is small, and the distribution of abnormal visual features is dispersed, such that the existing supervised detection methods cannot effectively detect undefined defects. At the same time, the normal texture of the aluminum profile surface presents non-uniform and non-periodic features, and this irregular distribution makes it difficult for classical reconstruction networks to accurately reconstruct the normal features, resulting in low performance of related unsupervised detection methods. Aiming at such problems, a feature-oriented reconstruction method of unsupervised surface-defect detection method for aluminum profiles is proposed. The aluminum profile image preprocessing stage uses techniques such as boundary extraction, background removal, and data normalization to process the original image and extract the image of the main part of the aluminum profile, which reduces the influence of irrelevant data features on the algorithm. The essential features learning stage precedes the feature-optimization module to eliminate the texture interference of the irregular distribution of the aluminum profile surface, and image blocks of the area images are reconstructed one by one to extract the features through the mask. The defect-detection stage compares the structural similarity of the feature images before and after the reconstruction, and comprehensively determines the detection results. The experimental results improve detection precision by 1.4% and the F1 value by 1.2% over the existing unsupervised methods, proving the effectiveness and superiority of the proposed method.

show abstract

Section: Related Work 21 Deep Learning Detection Methodsmentioning

confidence: 99%

A Feature-Oriented Reconstruction Method for Surface-Defect Detection on Aluminum Profiles

Tang,

Zhang,

Jin

et al. 2023

Applied Sciences

View full text Add to dashboard Cite

show abstract

“…Their work on AI and IoT for sustainable farming encapsulates the trend towards integrating sophisticated technologies in agriculture to enhance efficiency and environmental sustainability. Zhou and Yin [38] further this understanding by mapping the knowledge structure and trends in digital agriculture, providing a comprehensive overview of the field's evolution and future directions.…”

Section: In-depth Review Of Models Used For Multimodal Analysis Of Cr...mentioning

confidence: 99%

PAMICRM: Improving Precision Agriculture Through Multimodal Image Analysis for Crop Water Requirement Estimation Using Multidomain Remote Sensing Data Samples

Munaganuri,

Rao

2024

IEEE Access

View full text Add to dashboard Cite

The growing necessity for sustainable agriculture in the face of escalating global food demands and climate change underscores the importance of optimizing crop water usage. Current methodologies for estimating crop water requirements, primarily relying on traditional remote sensing and empirical models, face limitations in precision, adaptability to diverse crop types, and dynamic environmental conditions. This paper introduces an innovative machine learning model designed to accurately estimate crop water requirements and optimize irrigation scheduling using remote sensing data samples. The model's robustness stems from its sophisticated pre-processing technique, employing a Grey Wolf Optimized (GWO)-based Non-Local Means (NLM) algorithm for denoising images sourced from various remote sensing technologies. Subsequently, these refined images undergo a transformation into multidomain features through the integration of a Vision Transformer, Fourier, Entropy, Color Map, and Gabor Maps. This multi-faceted approach ensures a comprehensive analysis of the data, capturing intricate details pertinent to crop water requirements.A key innovation in our model is the implementation of the Dual Elephant Herding Optimization (DEHO) Process. This optimizer adeptly selects features by minimizing intra-class differences and maximizing interclass variance, thereby enhancing the model's discriminative capabilities. The selected features are then classified into distinct irrigation schedule classes using a Deep Dyna Q Graph Convolutional Network (DDQGCN). This network not only augments the efficiency of classifying images from diverse sources but also bolsters the model's capability to adapt to varying crop types and environmental conditions. The efficacy of the proposed model is further augmented by incorporating vector autoregressive moving average with eXogenous inputs (VARMAx) algorithms. This addition enables the conversion of output into temporal features, facilitating precise predictions of soil quality and the dynamic adjustment of irrigation schedules. Empirical testing of this model across different geographies on mango, rice, and cotton crops demonstrated significant improvements over existing methods. The model exhibited an 8.5% increase in precision, 8.3% in accuracy, 4.9% in recall, 5.5% in Area Under the Curve (AUC), and 4.5% in specificity, alongside a 3.5% reduction in response delays. Particularly noteworthy is the model's enhanced pre-emption capabilities, reflected in a 2.9% higher precision of pre-emption, 2.5% greater accuracy, 1.5% increased recall, 2.4% higher AUC of pre-emption, 1.5% improved specificity, and a 1.9% reduction in delay compared to existing methodologies.

show abstract

“…Explainability inside un-supervised methods is quite adopted, mainly based on anomaly maps [50], [199], [200], [204], [205], to perform the final segmentation through a threshold mechanism [3], [192], [208], [210], [212].…”

Section: D: Exploiting Activation Anomaly Mapsmentioning

confidence: 99%

Deep Learning for Automatic Vision-Based Recognition of Industrial Surface Defects: A Survey

et al. 2023

View full text Add to dashboard Cite

Automatic vision-based inspection systems have played a key role in product quality assessment for decades through the segmentation, detection, and classification of defects. Historically, machine learning frameworks, based on hand-crafted feature extraction, selection, and validation, counted on a combined approach of parameterized image processing algorithms and explicated human knowledge. The outstanding performance of deep learning (DL) for vision systems, in automatically discovering a feature representation suitable for the corresponding task, has exponentially increased the number of scientific articles and commercial products aiming at industrial quality assessment. In such a context, this article reviews more than 220 relevant articles from the related literature published until February 2023, covering the recent consolidation and advances in the field of fully-automatic DL-based surface defects inspection systems, deployed in various industrial applications. The analyzed papers have been classified according to a bi-dimensional taxonomy, that considers both the specific defect recognition task and the employed learning paradigm. The dependency on large and high-quality labeled datasets and the different neural architectures employed to achieve an overall perception of both well-visible and subtle defects, through the supervision of fine or/and coarse data annotations have been assessed. The results of our analysis highlight a growing research interest in defect representation power enrichment, especially by transferring pre-trained layers to an optimized network and by explaining the network decisions to suggest trustworthy retention or rejection of the products being evaluated.

show abstract

Defect Detection Method Based on Knowledge Distillation

Cited by 12 publications

References 52 publications

A Feature-Oriented Reconstruction Method for Surface-Defect Detection on Aluminum Profiles

A Feature-Oriented Reconstruction Method for Surface-Defect Detection on Aluminum Profiles

PAMICRM: Improving Precision Agriculture Through Multimodal Image Analysis for Crop Water Requirement Estimation Using Multidomain Remote Sensing Data Samples

Deep Learning for Automatic Vision-Based Recognition of Industrial Surface Defects: A Survey

Contact Info

Product

Resources

About