Zhen Yang

Defect detection in multimedia data plays a pivotal role in industrial manufacturing. However, existing methods are primarily designed for closed-world scenarios and can only identify defect classes in the training data, limiting their ability to effectively detect unknown class defects that arise during production. To address this critical limitation, we propose a novel approach by introducing industrial defect open set recognition (IDOSR), which overcomes the challenge of recognizing unknown defect classes. Furthermore, to tackle the issues of limited training samples and subtle inter-class differences in IDOSR, we present a high-frequency feature enhancement open set recognition (HFFE-OSR) method. Specifically, HFFE-OSR employs a high-frequency structural feature fusion enhancement strategy to meticulously extract and fuse defect-related high-frequency structural features. This enables the network to comprehensively learn defect target representations even under limited training samples, resulting in robust feature extraction for known classes, thereby improving the discriminability between known and unknown classes and addressing the difficulty of distinguishing between them. Additionally, a class mutual information constraint strategy is introduced to measure and reduce the mutual information among defect features from different classes. This ensures the independence of defect features across known classes, further enhancing their discriminability and significantly improving recognition performance for known classes. Extensive experiments demonstrate that the proposed method significantly outperforms state-of-the-art OSR methods on ID-OSD and MVTec datasets, achieving improvements of at least 7% in accuracy (ACC), 22% in F1 score, and 10% in AUROC, highlighting the effectiveness of our approach in industrial defect detection.

Medical image target segmentation is critical for clinical diagnosis and treatment, yet mainstream methods perform poorly on small medical targets due to inefficient sparse feature extraction and severe background interference. To tackle this issue, we propose Radiation Attenuation Characteristics Network (RAC-Net) inspired by the X-ray attenuation physical prior in CT imaging, aiming to boost small target segmentation accuracy for early lesion detection and localization. Specifically, RAC-Net integrates two tailored innovative modules: the Radiation Attenuation Salient Feature Extraction (RASFE) module leverages X-ray attenuation principles to extract small target Gaussian saliency features and suppress background noise, enhancing sparse feature representation; the Partial Volume Artifact Edge Enhancement (PVA) module embedded with CT Hounsfield Unit value prior fuses learnable neighborhood weights and physical prior weights, alleviating feature distortion and edge blurring caused by partial volume artifacts, while improving model robustness via clinical medical priors. Extensive experiments validate that RAC-Net achieves state-of-the-art performance in medical small target segmentation, with 18% DICE and 19% IOU improvements over advanced existing methods, providing solid support for early disease diagnosis and intervention.

Industrial defect detection is crucial in the field of industrial production. Recently, industrial defect detection methods based deep learning have been proven to effectively address most industrial defect detection tasks. However, these methods are unable to detect and identify defects of both known and new categories simultaneously. In industrial settings, the emergence of new types of defects is a common occurrence, rendering traditional closed set defect detection approaches inadequate for identifying these new defects. Therefore, to tackle this limitation, we introduce the concept of open set defect detection in industrial contexts, specifically formulated to recognize new types of defects. Furthermore, to address the challenges of limited training samples and small differences between defect classes in industrial defect open set detection, we propose a novel two-stage end-to-end methodology, termed Open Set Industrial Defect Detection (Open-IndDet). The core idea of Open-IndDet is to increase the discrimination boundary between known and new classes of defects by enhancing defect structural features and extracting class-unique robust features. Specifically, in the Open-IndDet, a class-unique robust feature extraction open set recognition strategy is proposed. This strategy fuses the structural features of defects in the defect confidence image with the defect features in the original image, thereby enhancing the expression of the structural features of defects and achieving full learning of defect features under limited training samples. In addition, this strategy extracts robust class-unique features by constraining the uniqueness of features within defect classes and enhancing the entropy of inter-class distribution differences. Such a strategy can increase the differences between the features of defects of different classes, achieving accurate classification of known class and identification of unknown class. The experimental results show that in our constructed standard open set detection dataset ID-OSD and public dataset MVTec, the proposed Open-IndDet achieved the best performance compared to existing advanced open set recognition methods.

Dense pixel-wise labeling of large-scale remote sensing images (RSI) is very time-consuming, while sparse labels (i.e., points, scribbles, or blocks) can be an efficient way to reduce labeling costs. Most existing sparse label-based methods adopt only one type of label for image segmentation, which cannot reflect the complex land covers in the RSI for training the model, thus leading to inferior segmentation performance. We observe that land covers with different shapes and complexity can be optimally represented by different sparse labels. Inspired by this observation, we propose a novel sparse labeling framework, termed Hybrid Sparse Labeling (HSLabeling), for large-scale RSI segmentation. Our HSLabeling can adaptively select the optimal hybrid sparse labels for different land covers, according to labeling cost and segmentation contribution of different sparse labels. Specifically, we first propose a label segmentation contribution information estimation module that estimates the information of different sparse labels according to the diversity and shape of land covers. After that, we propose an Optimal Hybrid Labeling Strategy (OHLS) to assign optimal types of labels for different land covers. In the OHLS, label assignment is formulated as an optimization problem that trades off label segmentation contribution information and labeling cost. We employ the greedy algorithm to efficiently solve the optimization problem and adaptively assign labels for varied land covers. Extensive experiments on three large-scale RSI datasets have demonstrated that our HSLabeling achieves almost fully supervised performance with extremely low labeling costs. In addition, compared with the single type sparse label, HSLabeling can also utilize much lower labeling costs to obtain the same performance. The source code is available at https://github.com/linjiaxing99/HSLabeling.

The small scale and weak intensity of infrared (IR) small targets result in extremely sparse target features, making small target detection in complex environments extremely susceptible to strong background noise interference. Most existing detection methods only focus on sparse feature extraction of targets, while ignoring the problem of strong background clutter interference, which leads to significant limitations in detection performance in complex scenes. Therefore, to address this issue, we propose a new IR small target detection network based on high- and low-frequency semantic reconstruction, namely the HLSR-net. In the HLSR-net, we draw inspiration from the frequency characteristics differences between targets and backgrounds in IR images, and use the proposed high- and low-frequency semantic reconstruction strategy to suppress background clutter interference, effectively extracting sparse features of targets. First, in the latent feature space of the input image, high-frequency target and low-frequency background features are extracted separately through fast Fourier frequency transformation. Second, by utilizing the difference in frequency characteristics between small targets and background clutter, target enhancement and background suppression are respectively carried out through the designed high- and low-frequency reconstruction networks. Finally, to further enrich the feature expression of the target, we designed a fusion output strategy to achieve detection of small targets. Sufficient experimental results indicate that the proposed HLSR-net achieves the performance of SOTA on the benchmark dataset, with an accuracy improvement of 6%.

In recent years, remote sensing recognition technology has found extensive applications in diverse fields, such as modern agriculture, forest management, urban planning, natural resource management, and disaster monitoring. However, the existing remote sensing recognition tasks face significant challenges because of the complex and ever-changing observation environment and the rapid growth of observation classes. The detection performance of existing closed-set recognition methods (where the test set does not contain unknown classes) is greatly limited. Therefore, numerous remote sensing open-set recognition (RSOSR) methods have been proposed to cope with more demanding but practical scenarios in the open world, including scenes or targets with unknown classes. Despite this, there is still a lack of comprehensive work on RSOSR technology. This article presents a comprehensive review of existing RSOSR technologies, covering relevant definitions, model principles, evaluation standards, and method comparisons. We then identify and discuss the limitations of current RSOSR technologies while highlighting promising research directions.

Cigarette defect detection is important in industrial production. Existing methods extract features for defect detection manually or using deep learning. However, due to the small size of cigarette defects, these methods are unable to effectively extract discriminative features, limiting detection performance. Hence, a deep learning‐based method called significant feature‐guided cigarette defect detection (SFGCD) is proposed, which combines saliency feature extraction methods with deep learning to enhance feature representation and improve detection. First, edge saliency features are extracted using the proposed target gradient saliency feature extraction (GSFE) strategy. Then, a dense multi‐level feature fusion network is designed to combine the original features with the saliency features obtained from the target gradient saliency feature extraction strategy. This network enriches feature representation and improves detection by fusing original and saliency features at different levels and scales. Experimental results demonstrate that the proposed method achieves a higher accuracy of 0.02 mean average precision (MAP) value and a detection speed of 5 frames per second (FPS) on the authors' own labeled cigarette defect dataset compared to existing state‐of‐the‐art methods.

The success of deep learning methods heavily relies on the availability of adequate samples. However, in the task of infrared small target detection (ISTD), the lack of high-quality training samples is a challenging problem due to the confidentiality of the application field and the difficulty of labeling. This limitation often leads to suboptimal detection performance of convolutional neural networks (CNNs). To address this challenge, we propose an adaptive labeling strategy and an ISTD network called the multi-criteria decision network (MCDNet) to achieve higher-quality sample labeling and more accurate detection results. In the adaptive labeling strategy, we propose a second-order differential autocorrelation method to determine the size of fuzzy edge targets accurately. In addition, we introduce local backgrounds to enhance the saliency information in the labels and improve the richness and contrast of training label content. To obtain accurate and robust detection results with limited target feature information, we design MCDNet. In particular, we propose a multi-criteria decision method that can combine the CNN decisions and the infrared small target prior saliency decisions through weighted fusion, and set decision weights based on the importance of different decision criteria in the decision-making process. This method can integrate the advantages of both the CNN decisions and the prior saliency decisions, avoid the one-sidedness of a single criterion, and improve the reliability and stability of the decision-making. The experimental results indicate that our method has a higher accuracy compared to other contrastive methods.

Robust infrared dim small target detection under sparse features is a difficult problem in the field of small target detection. Therefore, in order to solve this problem, this paper proposes an infrared dim small target detection network based on multi‐scale Gaussian significance and attention feature fusion (MGAF)‐net. In MGAF‐net, MGAF module is proposed to extract multi‐scale Gaussian saliency features of small targets. At the same time, in order to better fuse the multi‐scale Gaussian saliency features of the target, the attention mechanism is introduced in the MGAF module to enrich the feature expression of the target. The experimental results show that the proposed detection method has higher detection accuracy and higher detection speed than the existing advanced infrared dim small target detection methods.