Inter-class Similarity Research Articles

Group benefits instance for data purification

Manually annotating datasets for training deep models is very labor-intensive and time-consuming. To overcome such inferiority, directly leveraging web images to conduct training data becomes a natural choice. Nevertheless, the presence of label noise in web data usually degrades the model performance. Existing methods for combating label noise are typically designed and tested on synthetic noisy datasets. However, they tend to fail to achieve satisfying results on real-world noisy datasets. To this end, we propose a method named GRIP to alleviate the noisy label problem for both synthetic and real-world datasets. Specifically, GRIP utilizes a group regularization strategy that estimates class soft labels to improve noise-robustness. Soft label supervision reduces overfitting on noisy labels and learns inter-class similarities to benefit classification. Furthermore, an instance purification operation globally identifies noisy labels by measuring the difference between each training sample and its class soft label. Through operations at both group and instance levels, our approach integrates the advantages of noise-robust and noise-cleaning methods and remarkably alleviates the performance degradation caused by noisy labels. Comprehensive experimental results on synthetic and real-world datasets demonstrate the superiority of GRIP over the existing state-of-the-art methods. The data and source code of this work have been made available at: https://github.com/NUST-Machine-Intelligence-Laboratory/GRIP.

Semantic-Oriented Feature Coupling Transformer for Vehicle Re-Identification in Intelligent Transportation System

More robust intelligent transportation systems including autonomous driving systems are in full flourish with the revolution of deep learning and the 6G wireless communication network. Vehicle Re-Identification, an indispensable branch of the intelligent transportation system, aims to retrieve specific vehicles captured from non-overlapping cameras. However, this is fundamentally challenging with the substantial inter-class similarity and substantial intra-class divergence. Embedding semantic information into vehicle re-identification task has gained ample interest, but the performance needs to be further improved. This work proposes a semantic-oriented feature coupling transformer (SOFCT) for vehicle re-identification as a solution. Specifically, the knowledge-based transformer is first embedded to model images with discriminative attributes. Second, original patches are divided into five semantic groups via semantics-patches coupling, and the feature extractions for different semantics are performed in the semantic feature extraction (SFE) transformer. Third, patch features are weighted via semantics-patches coupling in the patch feature weighting (PFW) transformer, the weighted feature is fed into subsequent encoders to excavate information. Finally, two groups of learnable semantics are embedded to automatically learn semantic features in the learnable semantic extraction (LSE) transformer. Experiments demonstrate that the proposed SOFCT method surpasses other state-of-the-arts with the mAP/Rank-1 of 80.7%/96.6%, 89.8%/84.5%, 86.4%/80.9%, and 84.3%/78.7% on VeRi776 and VehicleID.

Rice seed classification by hyperspectral imaging system: A real-world dataset and a credible algorithm

Rice is one of the three major crops in the world. The hyperspectral identification of rice seeds is of great significance to the development of agriculture and hence has made remarkable achievements. However, the lack of publicly available benchmark datasets with a large number of rice seeds hinders the generalization of models and limits the advancement of the field. To address this issue, this study proposes a publicly accessible dataset for rice seed classification by hyperspectral imaging system. The proposed dataset contains 6 categories of nearly 10,000 seeds each and is characterized by (1) a large scale on the number of seeds per category and the total seed number; (2) a greater adherence to real-world scenarios; and (3) high intra-class diversity and high inter-class similarity. Based on the proposed dataset, an Instance Difficulty-weighted K-Nearest Neighbor algorithm (IDKNN) is further proposed. It is a credible algorithm, emphasizing individual seed accuracy rather than overall performance. Additionally, we explore the effects of different Regions Of Interest (ROI). Experimental results constitute the baseline performance of the proposed dataset. Compared with 10 representative algorithms, the IDKNN algorithm achieves the best performance, especially on the ROI of the entire seed. The dataset can be downloaded from https://github.com/fliery/RSHI60T.

A deep integrative approach for diabetic retinopathy classification with synergistic channel-spatial and self-attention mechanism

Diabetic retinopathy (DR) is a prevalent eye disease that poses a significant risk of vision loss in individuals with diabetes. Accurate classification of DR is crucial for timely intervention and effective treatment. However, the classification task is challenging due to imbalanced datasets, small lesions, and inter-class similarity. This study proposes a novel deep integrative approach for DR classification, leveraging the strengths of residual blocks, channel-spatial attention mechanism (CSAM), and non-local blocks (NLB). The proposed architecture consists of a sequence of residual blocks followed by CSAM modules, which enhance feature discriminability at different scales in DR images. Additionally, the outputs from CSAM blocks are fed into an NLB to capture long-range dependencies, allowing the model to consider global context information. The proposed architecture primarily utilizes the strength of global features extracted from non-local blocks and fuses them to enrich the information representation. Experimental results demonstrate that the proposed method effectively addresses challenges in DR classification and shows superior performance in terms of computational time and improved accuracy compared to existing methods.

Multi-contrast learning-guided lightweight few-shot learning scheme for predicting breast cancer molecular subtypes.

Invasive gene expression profiling studies have exposed prognostically significant breast cancer subtypes: normal-like, luminal, HER-2 enriched, and basal-like, which is defined in large part by human epidermal growth factor receptor 2 (HER-2), progesterone receptor (PR), and estrogen receptor (ER). However, while dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has been generally employed in the screening and therapy of breast cancer, there is a challenging problem to noninvasively predict breast cancer molecular subtypes, which have extremely low-data regimes. In this paper, a novel few-shot learning scheme, which combines lightweight contrastive convolutional neural network (LC-CNN) and multi-contrast learning strategy (MCLS), is worthwhile to be developed for predicting molecular subtype of breast cancer in DCE-MRI. Moreover, MCLS is designed to construct One-vs-Rest and One-vs-One classification tasks, which addresses inter-class similarity among normal-like, luminal, HER-2 enriched, and basal-like. Extensive experiments demonstrate the superiority of our proposed scheme over state-of-the-art methods. Furthermore, our scheme is able to achieve competitive results on few samples due to joint LC-CNN and MCLS for excavating contrastive correlations of a pair of DCE-MRI.

LVONet: automatic classification model for large vessel occlusion based on the difference information between left and right hemispheres.

Objective.Stroke is a highly lethal condition, with intracranial vessel occlusion being one of its primary causes. Intracranial vessel occlusion can typically be categorized into four types, each requiring different intervention measures. Therefore, the automatic and accurate classification of intracranial vessel occlusions holds significant clinical importance for assessing vessel occlusion conditions. However, due to the visual similarities in shape and size among different vessels and variations in the degree of vessel occlusion, the automated classification of intracranial vessel occlusions remains a challenging task. Our study proposes an automatic classification model for large vessel occlusion (LVO) based on the difference information between the left and right hemispheres.Approach.Our approach is as follows. We first introduce a dual-branch attention module to learn long-range dependencies through spatial and channel attention, guiding the model to focus on vessel-specific features. Subsequently, based on the symmetry of vessel distribution, we design a differential information classification module to dynamically learn and fuse the differential information of vessel features between the two hemispheres, enhancing the sensitivity of the classification model to occluded vessels. To optimize the feature differential information among similar vessels, we further propose a novel cooperative learning loss function to minimize changes within classes and similarities between classes.Main results.We evaluate our proposed model on an intracranial LVO data set. Compared to state-of-the-art deep learning models, our model performs optimally, achieving a classification sensitivity of 93.73%, precision of 83.33%, accuracy of 89.91% and Macro-F1 score of 87.13%.Significance.This method can adaptively focus on occluded vessel regions and effectively train in scenarios with high inter-class similarity and intra-class variability, thereby improving the performance of LVO classification.

MSE-Net: A novel master–slave encoding network for remote sensing scene classification

Remote sensing scene (RSS) image classification plays a vital role in various fields such as urban planning and environmental protection. However, due to higher inter-class similarity and intra-class variability, achieving accurate classification for RSS images poses a considerable challenge for current convolutional neural networks (CNNs)-based and visual transformer (ViT)-based methods. To address these issues, this paper proposes a novel dual-encoding method named master–slave encoding network (MSE-Net) from two perspectives of feature extraction and fusion. The master encoder, based on ViT, extracts higher-level semantic features, while the slave encoder, based on CNN, captures relative lower-level spatial structure information. Secondly, to integrate feature information from the two encoders effectively, this paper further develop two fusion strategies. The first strategy involves the auxiliary enhancement units (AEUs), which eliminates semantic divergence between the two encoders, enhances spatial context awareness of the slave encoder and promotes effective feature learning. The interactive perception unit (IPU), as the second strategy, facilitates interaction and integration of the two encoders’ representations to extract more discriminative feature information. In addition, we conducted comparative experiments on four widely-used RSS datasets, including RSSCN7, SIRI-WHU, the aerial image dataset (AID) and NWPU-RESISC45 (NWPU45), to verify the effectiveness of MSE-Net. The experimental results demonstrate that MSE-Net achieved state-of-the-art (SOTA) performance across all the datasets.

Context-aware and local-aware fusion with transformer for medical image segmentation

Objective. Convolutional neural networks (CNNs) have made significant progress in medical image segmentation tasks. However, for complex segmentation tasks, CNNs lack the ability to establish long-distance relationships, resulting in poor segmentation performance. The characteristics of intra-class diversity and inter-class similarity in images increase the difficulty of segmentation. Additionally, some focus areas exhibit a scattered distribution, making segmentation even more challenging. Approach. Therefore, this work proposed a new Transformer model, FTransConv, to address the issues of inter-class similarity, intra-class diversity, and scattered distribution in medical image segmentation tasks. To achieve this, three Transformer-CNN modules were designed to extract global and local information, and a full-scale squeeze-excitation module was proposed in the decoder using the idea of full-scale connections. Main results. Without any pre-training, this work verified the effectiveness of FTransConv on three public COVID-19 CT datasets and MoNuSeg. Experiments have shown that FTransConv, which has only 26.98M parameters, outperformed other state-of-the-art models, such as Swin-Unet, TransAttUnet, UCTransNet, LeViT-UNet, TransUNet, UTNet, and SAUNet++. This model achieved the best segmentation performance with a DSC of 83.22% in COVID-19 datasets and 79.47% in MoNuSeg. Significance. This work demonstrated that our method provides a promising solution for regions with high inter-class similarity, intra-class diversity and scatter distribution in image segmentation.

Joint margin adaption and multiscale feature fusion for COVID-19 CT images segmentation

Segmenting COVID-19 from CT images remains a challenging task due to the structural characteristics, including complex and diverse infected regions, high inter-class similarity, and intra-class variability, which make it difficult to capture detailed semantic information such as edges and textures effectively. The present study introduces a novel methodology for segmenting COVID-19 CT images, consisting of three distinct stages. The proposed approach is primarily based on the margin adaptive deep supervised feature fusion network (MADFNet) architecture, which adopts an encoder-decoder framework. In this strategy, the segmentation outcomes obtained in the previous phase are utilized as supervised terms to facilitate the semantic segmentation process in the subsequent stage. In MADFNet, firstly, the channel edge detection and spatial edge detection modules capture the rich semantic information of edge context from different dimensions to obtain the fine texture information and semantic dependencies, respectively. Then, the encoder feature fusion module is designed to aggregate multiscale feature representations at the same level, facilitating information integration across different channels and enabling adaptive detection of structural features such as edges and textures. Finally, the decoder feature fusion module integrates multiscale feature dependencies from different levels to alleviate the semantic discrepancies of features at different levels and accurately locate lesion regions. The single-class segmentation Dice scores of MADFNet on the three experimental datasets are 0.831, 0.825, and 0.821, respectively, and the results of single-class as well as multi-class lesion segmentation on COVID-19 CT images validated the feasibility of our work.

Surface defect detection of sawn timbers based on efficient multilevel feature integration

Surface defect detection of sawn timber is a critical task to ensure the quality of wooden products. Current methods have challenges in considering detection accuracy and speed simultaneously, due to the complexity of defects and the massive length of sawn timbers. Specifically, there are scale variation, large intraclass difference and high interclass similarity in the defects, which reduce the detection accuracy. To overcome these challenges, we propose an efficient multilevel-feature integration network (EMINet) based on YOLOv5s. To obtain discriminative features of defects, the cross fusion module (CFM) is proposed to fully integrate the multilevel features of backbone. In the CFM, the local information aggregation is designed to enrich the detailed information of high-level features, and the global information aggregation is designed to enhance the semantic information of low-level features. Experimental results demonstrate that the proposed EMINet achieves better accuracy with fast speed compared with the state-of-the-art methods.

A discriminative multiscale feature extraction network for facial expression recognition in the wild

Driven by advancements in deep learning technologies, substantial progress has been achieved in the field of facial expression recognition over the past decade, while challenges remain brought about by occlusions, pose variations and subtle expression differences in unconstrained (wild) scenarios. Therefore, a novel multiscale feature extraction method is proposed in this paper, that leverages convolutional neural networks to simultaneously extract deep semantic features and shallow geometric features. Through the mechanism of channel-wise self-attention, prominent features are further extracted and compressed, preserving advantageous features for distinction and thereby reducing the impact of occlusions and pose variations on expression recognition. Meanwhile, inspired by the large cosine margin concept used in face recognition, a center cosine loss function is proposed to avoid the misclassification caused by the underlying interclass similarity and substantial intra-class feature variations in the task of expression recognition. This function is designed to enhance the classification performance of the network through making the distribution of samples within the same class more compact and that between different classes sparser. The proposed method is benchmarked against several advanced baseline models on three mainstream wild datasets and two datasets that present realistic occlusion and pose variation challenges. Accuracies of 89.63%, 61.82%, and 91.15% are achieved on RAF-DB, AffectNet and FERPlus, respectively, demonstrating the greater robustness and reliability of this method compared to the state-of-the-art alternatives in the real world.

NmODE-Unet: A Novel Network for Semantic Segmentation of Medical Images

Diabetic retinopathy is a prevalent eye disease that poses a potential risk of blindness. Nevertheless, due to the small size of diabetic retinopathy lesions and the high interclass similarity in terms of location, color, and shape among different lesions, the segmentation task is highly challenging. To address these issues, we proposed a novel framework named nmODE-Unet, which is based on the nmODE (neural memory Ordinary Differential Equation) block and U-net backbone. In nmODE-Unet, the shallow features serve as input to the nmODE block, and the output of the nmODE block is fused with the corresponding deep features. Extensive experiments were conducted on the IDRiD dataset, e_ophtha dataset, and the LGG segmentation dataset, and the results demonstrate that, in comparison to other competing models, nmODE-Unet showcases a superior performance.

GaitMPL: Gait Recognition With Memory-Augmented Progressive Learning.

Gait recognition aims at identifying the pedestrians at a long distance by their biometric gait patterns. It is inherently challenging due to the various covariates and the properties of silhouettes (textureless and colorless), which result in two kinds of pair-wise hard samples: the same pedestrian could have distinct silhouettes (intra-class diversity) and different pedestrians could have similar silhouettes (inter-class similarity). In this work, we propose to solve the hard sample issue with a Memory-augmented Progressive Learning network (GaitMPL), including Dynamic Reweighting Progressive Learning module (DRPL) and Global Structure-Aligned Memory bank (GSAM). Specifically, DRPL reduces the learning difficulty of hard samples by easy-to-hard progressive learning. GSAM further augments DRPL with a structure-aligned memory mechanism, which maintains and models the feature distribution of each ID. Experiments on two commonly used datasets, CASIA-B and OU-MVLP, demonstrate the effectiveness of GaitMPL. On CASIA-B, we achieve the state-of-the-art performance, i.e., 88.0% on the most challenging condition (Clothing) and 93.3% on the average condition, which outperforms the other methods by at least 3.8% and 1.4%, respectively. Code will be available at https://github.com/WhiteDOU/GaitMPL https://github.com/WhiteDOU/GaitMPL.

A NOVEL MULTIMODAL BIOMETRIC SYSTEM BASED ON DEEP FUSION OF ECG AND EAR

Due to their unique and measurable properties, biometric security systems are more reliable and secure than traditional ones. However, unimodal biometric systems suffer from various problems such as spoof attacks, non-universality, intra-class variances, inter-class similarities, and noisy data. To overcome these problems, multimodal biometric systems which utilize more trait features have emerged to efficiently authenticate the identity of the individuals in various real-world applications. Along the same line, this paper proposes a multimodal biometric system for human recognition based on deep features fusion of electrocardiograms (ECG) signals and ear images. The proposed system is hard to spoof compared to current systems as the ear biometric provides a fixed structure over an acceptable period of human life, and the ECG offers the characteristic of the person's liveness. It also applies a transfer-learning methodology to extract discriminative deep features by exploiting a pre-trained VGG-m Net model. Furthermore, to improve the efficiency of the proposed model’s training, augmentation techniques were utilized to further increase the size of the training data. A course of experiments has been conducted to assess the performance of the proposed approach for unimodal and multimodal biometric traits. The experimental results reveal that the proposed system achieves promising results and outperforms the unimodal of ECG and ear, and other state-of-the-art multimodal biometric systems.

Synthesizing Knowledge-Enhanced Features for Real-World Zero-Shot Food Detection.

Food computing brings various perspectives to computer vision like vision-based food analysis for nutrition and health. As a fundamental task in food computing, food detection needs Zero-Shot Detection (ZSD) on novel unseen food objects to support real-world scenarios, such as intelligent kitchens and smart restaurants. Therefore, we first benchmark the task of Zero-Shot Food Detection (ZSFD) by introducing FOWA dataset with rich attribute annotations. Unlike ZSD, fine-grained problems in ZSFD like inter-class similarity make synthesized features inseparable. The complexity of food semantic attributes further makes it more difficult for current ZSD methods to distinguish various food categories. To address these problems, we propose a novel framework ZSFDet to tackle fine-grained problems by exploiting the interaction between complex attributes. Specifically, we model the correlation between food categories and attributes in ZSFDet by multi-source graphs to provide prior knowledge for distinguishing fine-grained features. Within ZSFDet, Knowledge-Enhanced Feature Synthesizer (KEFS) learns knowledge representation from multiple sources (e.g., ingredients correlation from knowledge graph) via the multi-source graph fusion. Conditioned on the fusion of semantic knowledge representation, the region feature diffusion model in KEFS can generate fine-grained features for training the effective zero-shot detector. Extensive evaluations demonstrate the superior performance of our method ZSFDet on FOWA and the widely-used food dataset UECFOOD-256, with significant improvements by 1.8% and 3.7% ZSD mAP compared with the strong baseline RRFS. Further experiments on PASCAL VOC and MS COCO prove that enhancement of the semantic knowledge can also improve the performance on general ZSD. Code and dataset are available at https://github.com/LanceZPF/KEFS.

Smoke-Aware Global-Interactive Non-local Network for Smoke Semantic Segmentation.

Compared with other objects, smoke semantic segmentation (SSS) is more difficult and challenging due to some special characteristics of smoke, such as non-rigid, translucency, variable mode and so on. To achieve accurate positioning of smoke in real complex scenes and promote the development of intelligent fire detection, we propose a Smoke-Aware Global-Interactive Non-local Network (SAGINN) for SSS, which harness the power of both convolution and transformer to capture local and global information simultaneously. Non-local is a powerful means for modeling long-range context dependencies, however, friendliness to single-scale low-resolution features limits its potential to produce high-quality representations. Therefore, we propose a Global-Interactive Non-local (GINL) module, leveraging global interaction between multi-scale key information to improve the robustness of feature representations. To solve the interference of smoke-like objects, a Pyramid High-level Semantic Aggregation (PHSA) module is designed, where the learned high-level category semantics from classification aids model by providing additional guidance to correct the wrong information in segmentation representations at the image level and alleviate the inter-class similarity problem. Besides, we further propose a novel loss function, termed Smoke-aware loss (SAL), by assigning different weights to different objects contingent on their importance. We evaluate our SAGINN on extensive synthetic and real data to verify its generalization ability. Experimental results show that SAGINN achieves 83% average mIoU on the three testing datasets (83.33%, 82.72% and 82.94%) of SYN70K with an accuracy improvement of about 0.5%, 0.002 mMse and 0.805 Fβ on SMOKE5K, which can obtain more accurate location and finer boundaries of smoke, achieving satisfactory results on smoke-like objects.

I $^{2}$ RNN: An Incremental and Interpretable Recurrent Neural Network for Encrypted Traffic Classification

Traffic classification occupies a significant role in cybersecurity and network management. The widespread of encryption transmission protocols such as SSL/TLS has led to the dominance of deep learning based approaches. In cybersecurity, strong adversaries often complicate their strategies by constantly developing emerging attacks. Meanwhile, security practitioners desire to grasp the reasons for inference results. However, existing deep learning approaches lack efficient adaptation for incremental traffic types and often have less interpretability. In this paper, we propose I <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> RNN, an Incremental and Interpretable Recurrent Neural Network for encrypted traffic classification. The I <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> RNN proposes a novel propagation process to extract the sequence fingerprints from sessions with local robustness. Meanwhile, this proposal provides interpretability including time-series feature attribution and inter-class similarity portrait. Moreover, we design I <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> RNN in an incremental manner to adapt to emerging traffic types. The I <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> RNN only needs to train an additional set of parameters for the newly added traffic type rather than retraining the whole model with the entire dataset. Extensive experimental results show that our I <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> RNN can achieve remarkable performance in traffic classification, incremental learning, and model interpretability. Compared with other local interpretability methods, our I <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> RNN exhibits excellent stability, robustness, and effectiveness in the interpretation of network traffic data.

Deep Augmented Metric Learning Network for Prostate Cancer Classification in Ultrasound Images.

Prostate cancer screening often relies on cost-intensive MRIs and invasive needle biopsies. Transrectal ultrasound imaging, as a more affordable and non-invasive alternative, faces the challenge of high inter-class similarity and intra-class variability between benign and malignant prostate cancers. This complexity requires more stringent differentiation of subtle features for accurate auxiliary diagnosis. In response, we introduce the novel Deep Augmented Metric Learning (DAML) network, specifically tailored for ultrasound-based prostate cancer classification. The DAML network represents a significant innovation in the metric learning space, introducing the Semantic Differences Mining Strategy (SDMS) to effectively discern and represent subtle differences in prostate ultrasound images, thereby enhancing tumor classification accuracy. Additionally, the DAML network strategically addresses class variability and limited sample sizes by combining the Linear Interpolation Augmentation Strategy (LIAS) and Permutation-Aided Reconstruction Loss (PARL). This approach enriches feature representation and introduces variability with straightforward structures, mirroring the efficacy of advanced sample generation techniques. We carried out comprehensive empirical assessments of the DAML model by testing its key components against a range of models, ensuring its effectiveness. Our results demonstrate the enhanced performance of the DAML model, achieving classification accuracies of 0.857 and 0.888 for benign and malignant cancers, respectively, underscoring its effectiveness in prostate cancer classification via medical imaging.

Fault Diagnosis of Complex Industrial Systems Based on Multi-Granularity Dictionary Learning and Its Application

Nowadays, the intelligent fault diagnosis problem of modern industrial systems has received increasing attention. However, with the increasing scale of industrial systems, the same category of faulty data often contains multiple working conditions, which leads to the multi-granularity of faulty data and the increasing complexity of data distribution. The possible existence of fine-granularity inter-class similarities in multi-granularity data can interfere with coarse-granularity recognition tasks, which brings challenges to data-driven fault diagnosis tasks. To solve the above issue, a fault diagnosis method based on multi-granularity dictionary learning is proposed in this paper. Specifically, the proposed method integrates prior knowledge into the dictionary learning framework by introducing the concept of “centroid atom”, which embeds the multi-granularity structure in the dictionary and improves the discrimination of dictionary atoms. In order to validate the proposed approach, experiments are conducted on a numerical simulation case and a zinc smelting roasting process. Compared with some state-of-the-art methods, the proposed approach performs better in the classification accuracy which shows that it can make full use of the knowledge provided by the multi-granularity structure while it is difficult to be modeled by the other methods. Therefore, it can be applied to fault diagnosis tasks more precisely and efficiently. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —Motivated by the fact that industrial faults often occur in different modes, and different faults may occur in the same mode, the fault data often has multi-granularity characteristics, this paper proposes a multi-granularity dictionary learning method for complex industrial fault diagnosis. The proposed method can learn the multi-granularity structure in the data, and the obtained model contains both coarse-grained and fine-grained information, which effectively improves the fault diagnosis accuracy. Intensive experimental results show that the proposed method performs better than some state-of-the-art methods, it can make full use of the knowledge provided by the multi-granularity structure, while it is difficult to be modeled by others. Above all, it is verified as suitable for fault diagnosis of real industrial systems.

MDKLoss: Medicine domain knowledge loss for skin lesion recognition.

Methods based on deep learning have shown good advantages in skin lesion recognition. However, the diversity of lesion shapes and the influence of noise disturbances such as hair, bubbles, and markers leads to large intra-class differences and small inter-class similarities, which existing methods have not yet effectively resolved. In addition, most existing methods enhance the performance of skin lesion recognition by improving deep learning models without considering the guidance of medical knowledge of skin lesions. In this paper, we innovatively construct feature associations between different lesions using medical knowledge, and design a medical domain knowledge loss function (MDKLoss) based on these associations. By expanding the gap between samples of various lesion categories, MDKLoss enhances the capacity of deep learning models to differentiate between different lesions and consequently boosts classification performance. Extensive experiments on ISIC2018 and ISIC2019 datasets show that the proposed method achieves a maximum of 91.6% and 87.6% accuracy. Furthermore, compared with existing state-of-the-art loss functions, the proposed method demonstrates its effectiveness, universality, and superiority.