Search

As a fundamental and challenging task in bridging language and vision domains, Image-Text Retrieval (ITR) aims at searching for the target instances that are semantically relevant to the given query from the other modality, and its key challenge is to measure the semantic similarity across different modalities. Although significant progress has been achieved, existing approaches typically suffer from two major limitations: (1) It hurts the accuracy of the representation by directly exploiting the bottom-up attention based region-level features where each region is equally treated. (2) It limits the scale of negative sample pairs by employing the mini-batch based end-to-end training mechanism. To address these limitations, we propose a Unified Semantic Enhancement Momentum Contrastive Learning (USER) method for ITR. Specifically, we delicately design two simple but effective Global representation based Semantic Enhancement (GSE) modules. One learns the global representation via the self-attention algorithm, noted as Self-Guided Enhancement (SGE) module. The other module benefits from the pre-trained CLIP module, which provides a novel scheme to exploit and transfer the knowledge from an off-the-shelf model, noted as CLIP-Guided Enhancement (CGE) module. Moreover, we incorporate the training mechanism of MoCo into ITR, in which two dynamic queues are employed to enrich and enlarge the scale of negative sample pairs. Meanwhile, a Unified Training Objective (UTO) is developed to learn from mini-batch based and dynamic queue based samples. Extensive experiments on the benchmark MSCOCO and Flickr30K datasets demonstrate the superiority of both retrieval accuracy and inference efficiency. For instance, compared with the existing best method NAAF, the metric R@1 of our USER on the MSCOCO 5K Testing set is improved by 5% and 2.4% on caption retrieval and image retrieval without any external knowledge or pre-trained model while enjoying over 60 times faster inference speed. Our source code will be released at https://github.com/zhangy0822/USER. [ABSTRACT FROM AUTHOR]

Graph-based clustering plays an important role in the clustering area. Recent studies about graph neural networks (GNN) have achieved impressive success on graph-type data. However, in general clustering tasks, the graph structure of data does not exist such that GNN can not be applied to clustering directly and the strategy to construct a graph is crucial for performance. Therefore, how to extend GNN into general clustering tasks is an attractive problem. In this paper, we propose a graph auto-encoder for general data clustering, AdaGAE, which constructs the graph adaptively according to the generative perspective of graphs. The adaptive process is designed to induce the model to exploit the high-level information behind data and utilize the non-euclidean structure sufficiently. Importantly, we find that the simple update of the graph will result in severe degeneration, which can be concluded as better reconstruction means worse update. We provide rigorous analysis theoretically and empirically. Then we further design a novel mechanism to avoid the collapse. Via extending the generative graph models to general type data, a graph auto-encoder with a novel decoder is devised and the weighted graphs can be also applied to GNN. AdaGAE performs well and stably in different scale and type datasets. Besides, it is insensitive to the initialization of parameters and requires no pretraining. [ABSTRACT FROM AUTHOR]

Linear discriminant analysis (LDA) has been proven to be effective in dimensionality reduction. However, the performance of LDA depends on the consistency assumption of the global structure and the local structure. Some work extended LDA along this line of research and proposed local formulations of LDA. Unfortunately, the learning scheme of these algorithms is suboptimal in that the intrinsic relationship between data points is pre-learned in the original space, which is usually affected by the noise and redundant features. Besides, the time cost is relatively high. To alleviate these drawbacks, we propose a Fast Locality Discriminant Analysis framework (FLDA), which has three advantages: (1) It can divide a non-Gaussian distribution class into many sub-blocks that obey Gaussian distributions by using the anchor-based strategy. (2) It captures the manifold structure of data by learning the fuzzy membership relationship between data points and the corresponding anchor points, which can reduce computation time. (3) The weights between data points and anchor points are adaptively updated in the subspace where the irrelevant information and the noise in high-dimensional space have been effectively suppressed. Extensive experiments on toy data sets, UCI benchmark data sets and imbalanced data sets demonstrate the efficiency and effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

In this article, we focus on the unsupervised multiview feature selection, which tries to handle high-dimensional data in the field of multiview learning. Although some graph-based methods have achieved satisfactory performance, they ignore the underlying data structure across different views. Besides, their predefined Laplacian graphs are sensitive to the noises in the original data space and fail to obtain the optimal neighbor assignment. To address the above problems, we propose a novel unsupervised multiview feature selection model based on graph learning, and the contributions are three-fold: 1) during the feature selection procedure, the consensus similarity graph shared by different views is learned. Therefore, the proposed model can reveal the data relationship from the feature subset; 2) a reasonable rank constraint is added to optimize the similarity matrix to obtain more accurate information; and 3) an autoweighted framework is presented to assign view weights adaptively, and an effective alternative iterative algorithm is proposed to optimize the problem. Experiments on various datasets demonstrate the superiority of the proposed method compared to the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The location information of road and lane lines is the supremely important thing for the automatic drive and auxiliary drive. The detection accuracy of these two elements dramatically affects the reliability and practicality of the whole system. In real applications, the traffic scene can be very complicated, which makes it particularly challenging to obtain the precise location of road and lane lines. Commonly used deep learning-based object detection models perform pretty well on the lane line and road detection tasks, but they still encounter false detection and missing detection frequently. Besides, existing convolution neural network (CNN) structures only pay attention to the information flow between layers, while it cannot fully utilize the spatial information inside the layers. To address those problems, we propose an attention-based spatial segmentation network for traffic scene understanding. We use the convolutional attention module to improve the network’s understanding capacity of spatial location distribution. Spatial CNN (SCNN) obtains through the information flow within one single convolutional layer and improves the spatial relationship modeling ability of the network. The experimental results demonstrate that this method effectively improves the neural network’s application ability of the spatial information, thereby improving the effect of traffic scene understanding. Furthermore, a pixel-level road segmentation dataset called NWPU Road Dataset is built to help improve the process of traffic scene understanding. [ABSTRACT FROM AUTHOR]

Few-shot learning (FSL) for human–object interaction (HOI) aims at recognizing various relationships between human actions and surrounding objects only from a few samples. It is a challenging vision task, in which the diversity and interactivity of human actions result in great difficulty to learn an adaptive classifier to catch ambiguous interclass information. Therefore, traditional FSL methods usually perform unsatisfactorily in complex HOI scenes. To this end, we propose dynamic graph-in-graph networks (DGIG-Net), a novel graph prototypes framework to learn a dynamic metric space by embedding a visual subgraph to a task-oriented cross-modal graph for few-shot HOI. Specifically, we first build a knowledge reconstruction graph to learn latent representations for HOI categories by reconstructing the relationship among visual features, which generates visual representations under the category distribution of every task. Then, a dynamic relation graph integrates both reconstructible visual nodes and dynamic task-oriented semantic information to explore a graph metric space for HOI class prototypes, which applies the discriminative information from the similarities among actions or objects. We validate DGIG-Net on multiple benchmark datasets, on which it largely outperforms existing FSL approaches and achieves state-of-the-art results. [ABSTRACT FROM AUTHOR]

Linear discriminant analysis (LDA) is a well-known technique for supervised dimensionality reduction and has been extensively applied in many real-world applications. LDA assumes that the samples are Gaussian distributed, and the local data distribution is consistent with the global distribution. However, real-world data seldom satisfy this assumption. To handle the data with complex distributions, some methods emphasize the local geometrical structure and perform discriminant analysis between neighbors. But the neighboring relationship tends to be affected by the noise in the input space. In this research, we propose a new supervised dimensionality reduction method, namely, locality adaptive discriminant analysis (LADA). In order to directly process the data with matrix representation, such as images, the 2-D LADA (2DLADA) is also developed. The proposed methods have the following salient properties: 1) they find the principle projection directions without imposing any assumption on the data distribution; 2) they explore the data relationship in the desired subspace, which contains less noise; and 3) they find the local data relationship automatically without the efforts for tuning parameters. The performance of dimensionality reduction shows the superiorities of the proposed methods over the state of the art. [ABSTRACT FROM AUTHOR]

Scene graph generation (SGG) is built on top of detected objects to predict object pairwise visual relations for describing the image content abstraction. Existing works have revealed that if the links between objects are given as prior knowledge, the performance of SGG is significantly improved. Inspired by this observation, in this article, we propose a relation regularized network (R2-Net), which can predict whether there is a relationship between two objects and encode this relation into object feature refinement and better SGG. Specifically, we first construct an affinity matrix among detected objects to represent the probability of a relationship between two objects. Graph convolution networks (GCNs) over this relation affinity matrix are then used as object encoders, producing relation-regularized representations of objects. With these relation-regularized features, our R2-Net can effectively refine object labels and generate scene graphs. Extensive experiments are conducted on the visual genome dataset for three SGG tasks (i.e., predicate classification, scene graph classification, and scene graph detection), demonstrating the effectiveness of our proposed method. Ablation studies also verify the key roles of our proposed components in performance improvement. [ABSTRACT FROM AUTHOR]

Domain adaptation leverages rich knowledge from a related source domain so that it can be used to perform tasks in a target domain. For more knowledge to be obtained under relaxed conditions, domain adaptation methods have been widely used in pattern recognition and image classification. However, most of the existing domain adaptation methods only consider how to minimize different distributions of the source and target domains, which neglects what should be transferred for a specific task and suffers negative transfer by distribution outliers. To address these problems, in this paper, we propose a novel domain adaptation method called weighted correlation embedding learning (WCEL) for image classification. In the WCEL approach, we seamlessly integrated correlation learning, graph embedding, and sample reweighting into a unified learning model. Specifically, we extracted the maximum correlated features from the source and target domains for image classification tasks. In addition, two graphs were designed to preserve the discriminant information from interclass samples and neighborhood relations in intraclass samples. Furthermore, to prevent the negative transfer problem, we developed an efficient sample reweighting strategy to predict the target with different confidence levels. To verify the performance of the proposed method in image classification, extensive experiments were conducted with several benchmark databases, verifying the superiority of the WCEL method over other state-of-the-art domain adaptation algorithms. [ABSTRACT FROM AUTHOR]

Recently, crowd counting draws much attention on account of its significant meaning in congestion control, public safety, and ecological surveys. Although the performance is improved dramatically due to the development of deep learning, the scales of these networks also become larger and more complex. Moreover, a large model also entails more time to train for better performance. To tackle these problems, this article first constructs a lightweight model, which is composed of an image feature encoder and a simple but effective decoder, called the pixel shuffle decoder (PSD). PSD ends with a pixel shuffle operator, which can display more density information without increasing the number of convolutional layers. Second, a density-aware curriculum learning (DCL) training strategy is designed to fully tap the potential of crowd counting models. DCL gives each predicted pixel a weight to determine its predicting difficulty and provides guidance on obtaining better generalization. Experimental results exhibit that PSD can achieve outstanding performance on most mainstream datasets while training under the DCL training framework. Besides, we also conduct some experiments about adopting DCL on existing typical crowd counters, and the results show that they all obtain new better performance than before, which further validates the effectiveness of our method. [ABSTRACT FROM AUTHOR]

Unsupervised feature selection (UFS) aims to remove the redundant information and select the most representative feature subset from the original data, so it occupies a core position for high-dimensional data preprocessing. Many proposed approaches use self-expression to explore the correlation between the data samples or use pseudolabel matrix learning to learn the mapping between the data and labels. Furthermore, the existing methods have tried to add constraints to either of these two modules to reduce the redundancy, but no prior literature embeds them into a joint model to select the most representative features by the computed top ranking scores. To address the aforementioned issue, this article presents a novel UFS method via a convex non-negative matrix factorization with an adaptive graph constraint (CNAFS). Through convex matrix factorization with adaptive graph constraint, it can dig up the correlation between the data and keep the local manifold structure of the data. To our knowledge, it is the first work that integrates pseudo label matrix learning into the self-expression module and optimizes them simultaneously for the UFS solution. Besides, two different manifold regularizations are constructed for the pseudolabel matrix and the encoding matrix to keep the local geometrical structure. Eventually, extensive experiments on the benchmark datasets are conducted to prove the effectiveness of our method. The source code is available at: https://github.com/misteru/CNAFS. [ABSTRACT FROM AUTHOR]

In this article, we propose a novel feature selection approach via explicitly addressing the long-standing subspace sparsity issue. Leveraging $\ell _{2,1}$ -norm regularization for feature selection is the major strategy in existing methods, which, however, confronts sparsity limitation and parameter-tuning trouble. To circumvent this problem, employing the $\ell _{2,0}$ -norm constraint to improve the sparsity of the model has gained more attention recently whereas, optimizing the subspace sparsity constraint is still an unsolved problem, which only can acquire an approximate solution and without convergence proof. To address the above challenges, we innovatively propose a novel subspace sparsity discriminative feature selection (S2DFS) method which leverages a subspace sparsity constraint to avoid tuning parameters. In addition, the trace ratio formulated objective function extremely ensures the discriminability of selected features. Most important, an efficient iterative optimization algorithm is presented to explicitly solve the proposed problem with a closed-form solution and strict convergence proof. To the best of our knowledge, such an optimization algorithm of solving the subspace sparsity issue is first proposed in this article, and a general formulation of the optimization algorithm is provided for improving the extensibility and portability of our method. Extensive experiments conducted on several high-dimensional text and image datasets demonstrate that the proposed method outperforms related state-of-the-art methods in pattern classification and image retrieval tasks. [ABSTRACT FROM AUTHOR]

Small aerial object detection plays an important role in numerous computer vision tasks, including remote sensing, early warning systems, and visual tracking. Despite existing moving object detection techniques that can achieve reasonable results in normal size objects, they fail to distinguish the small objects from the dynamic background. To cope with this issue, a novel method is proposed for accurate small aerial object detection under different situations. Initially, the block segmentation is introduced for reducing frame information redundancy. Meanwhile, a random projection feature (RPF) is proposed for characterizing blocks into feature vectors. Subsequently, a moving direction estimation based on feature vectors is presented to measure the motions of blocks and filter out the major directions. Finally, variable search region clustering (VSRC), together with the color feature difference, is designed for extracting pixelwise targets from the remaining moving direction blocks. The comprehensive experiments demonstrate that our approach outperforms the level of state-of-the-art methods upon the integrity of small aerial objects, especially on the dynamic background and scale variation targets. [ABSTRACT FROM AUTHOR]

Exploiting the inner-shot and inter-shot dependencies is essential for key-shot based video summarization. Current approaches mainly devote to modeling the video as a frame sequence by recurrent neural networks. However, one potential limitation of the sequence models is that they focus on capturing local neighborhood dependencies while the high-order dependencies in long distance are not fully exploited. In general, the frames in each shot record a certain activity and vary smoothly over time, but the multi-hop relationships occur frequently among shots. In this case, both the local and global dependencies are important for understanding the video content. Motivated by this point, we propose a reconstructive sequence-graph network (RSGN) to encode the frames and shots as sequence and graph hierarchically, where the frame-level dependencies are encoded by long short-term memory (LSTM), and the shot-level dependencies are captured by the graph convolutional network (GCN). Then, the videos are summarized by exploiting both the local and global dependencies among shots. Besides, a reconstructor is developed to reward the summary generator, so that the generator can be optimized in an unsupervised manner, which can avert the lack of annotated data in video summarization. Furthermore, under the guidance of reconstruction loss, the predicted summary can better preserve the main video content and shot-level dependencies. Practically, the experimental results on three popular datasets (i.e., SumMe, TVsum and VTW) have demonstrated the superiority of our proposed approach to the summarization task. [ABSTRACT FROM AUTHOR]

In this paper, a novel low-rank structural model is proposed for segmenting data drawn from a high-dimensional space. Our method is based on the fact that all groups clustered from a high-dimensional dataset are distributed in multiple low-rank subspaces. In general, it’s a very difficult task to find the low-rank structures hidden in data. Different from the classical sparse subspace clustering (SSC) and low-rank representation (LRR) which all take two steps including building the affinity matrix and spectral clustering, we introduce a new rank constraint into our model. This constraint allows our model to learn a subspace indicator which can capture different clusters directly from the data without any postprocessing. To further approximate the rank constraint, a piecewise function is utilized as the relaxing item for the proposed model. Besides, under the subspace indicator constraints, the integer programming problem is avoided, which makes our algorithm more efficient and scalable. In addition, we prove the convergence of the proposed algorithm in theory and further discuss the general case in which subspaces don’t pass through the origin. Experiment results on both synthetic and real-world datasets demonstrate that our algorithm significantly outperforms the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Fuzzy K-Means (FKM) clustering is of great importance for analyzing unlabeled data. FKM algorithms assign each data point to multiple clusters with some degree of certainty measured by the membership function. In these methods, the fuzzy membership degree matrix is obtained based on the calculation of the distance between data points in the original space. However, this operation may lead to suboptimal results because of the influence of noises and redundant features. Besides, some FKM clustering methods ignore the importance of the weighting exponent. In this paper, we propose a novel FKM method called Fuzzy K-Means Clustering With Discriminative Embedding. Within this method, we simultaneously conduct dimensionality reduction along with fuzzy membership degree learning. To retain most information in the embedding subspace and improve the robustness of this method, principal component analysis is incorporated into our framework. An iterative optimization algorithm is proposed to solve the model. To validate the efficacy of the proposed method, we perform comprehensive analyses, including convergence behavior, parameter determination and computational complexity. Moreover, we also match a appropriate weighting exponent for each data set. Experimental results on benchmark data sets show that the proposed method is more discriminative and effective for clustering tasks. [ABSTRACT FROM AUTHOR]

Due to the exponential growth of multimedia data, multi-modal hashing as a promising technique to make cross-view retrieval scalable is attracting more and more attention. However, most of the existing multi-modal hashing methods either divide the learning process unnaturally into two separate stages or treat the discrete optimization problem simplistically as a continuous one, which leads to suboptimal results. Recently, a few discrete multi-modal hashing methods that try to address such issues have emerged, but they still ignore several important discrete constraints (such as the balance and decorrelation of hash bits). In this paper, we overcome those limitations by proposing a novel method named “Enhanced Discrete Multi-modal Hashing (EDMH)” which learns binary codes and hashing functions simultaneously from the pairwise similarity matrix of data, under the aforementioned discrete constraints. Although the model of EDMH looks a lot more complex than the other models for multi-modal hashing, we are actually able to develop a fast iterative learning algorithm for it, since the subproblems of its optimization all have closed-form solutions after introducing a couple of auxiliary variables. Our experimental results on three real-world datasets have revealed the usefulness of those previously ignored discrete constraints and demonstrated that EDMH not only performs much better than state-of-the-art competitors according to several retrieval metrics but also runs much faster than most of them. [ABSTRACT FROM AUTHOR]

Feature selection is one of the most important techniques to deal with the high-dimensional data for a variety of machine learning and data mining tasks, such clustering, classification, and retrieval, etc. Fuzziness is a widespread nature of data in nature human society. However, most existing feature selection methods ignore the existence of fuzziness in the data, resulting in sub-optimal feature subsets. To address the problem, we propose a novel unsupervised feature selection method, called Unsupervised Discriminative Projection for Feature Selection (UDPFS) to select discriminative features by conducting fuzziness learning and sparse learning, simultaneously. Specifically, we use projection matrix transform data as its low-dimensional representation, which are partitioned into clusters by using membership matrix with sparse constraint. In addition, $\ell _{2, 1}$ ℓ 2 , 1 -norm regularization is applied to the projection matrix. Then, a discriminative projection matrix with row sparse is obtained by perform fuzziness learning and sparse learning, simultaneously. An effective alternative optimization algorithm is proposed to solve the objective function. Evaluate experimental results on several real-world datasets show the effectiveness and superiority of the proposed unsupervised feature selection method. [ABSTRACT FROM AUTHOR]

Data clustering, which is to partition the given data into different groups, has attracted much attention. Recently various effective algorithms have been developed to tackle the task. Among these methods, non-negative matrix factorization (NMF) has been demonstrated to be a powerful tool. However, there are still some problems. First, the standard NMF is sensitive to noises and outliers. Although $\ell _{2,1}$ ℓ 2 , 1 norm based NMF improves the robustness, it is still affected easily by large noises. Second, for most graph regularized NMF, the performance highly depends on the initial similarity graph. Third, many graph-based NMF models perform the graph construction and matrix factorization in two separated steps. Thus the learned graph structure may not be optimal. To overcome the above drawbacks, we propose a robust bi-stochastic graph regularized matrix factorization (RBSMF) framework for data clustering. Specifically, we present a general loss function, which is more robust than the commonly used $L_2$ L 2 and $L_1$ L 1 functions. Besides, instead of keeping the graph fixed, we learn an adaptive similarity graph. Furthermore, the graph updating and matrix factorization are processed simultaneously, which can make the learned graph more appropriate for clustering. Extensive experiments have shown the proposed RBSMF outperforms other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Remote sensing image captioning (RSIC), which aims at generating a well-formed sentence for a remote sensing image, has attracted more attention in recent years. The general framework for RSIC is the encoder–decoder architecture containing two submodels of encoder and decoder. Although the significant performance is obtained, the encoder–decoder architecture is a black-box model with a lack of explainability. To overcome this drawback, in this article, we propose a new explainable word–sentence framework for RSIC. The proposed word–sentence framework consists of two parts: word extractor and sentence generator, where the former extracts the valuable words in the given remote sensing image, while the latter organizes these words into a well-formed sentence. The proposed framework decomposes RSIC into a word classification task and a word sorting task, which is more in line with human intuitive understanding. On the basis of the word–sentence framework, some ablation experiments are conducted on the three public RSIC data sets of Sydney-captions, UCM-captions, and RSICD to explore the specific and effective network structures. In order to evaluate the proposed word–sentence framework objectively, we further conduct some comparative experiments on these three data sets and achieve comparable results in comparison with the encoder–decoder-based methods. [ABSTRACT FROM AUTHOR]

Smoke has semi-transparency property leading to highly complicated mixture of background and smoke. Sparse or small smoke is visually inconspicuous, and its boundary is often ambiguous. These reasons result in a very challenging task of separating smoke from a single image. To solve these problems, we propose a Classification-assisted Gated Recurrent Network (CGRNet) for smoke semantic segmentation. To discriminate smoke and smoke-like objects, we present a smoke segmentation strategy with dual classification assistance. Our classification module outputs two prediction probabilities for smoke. The first assistance is to use one probability to explicitly regulate the segmentation module for accuracy improvement by supervising a cross-entropy classification loss. The second one is to multiply the segmentation result by another probability for further refinement. This dual classification assistance greatly improves performance at image level. In the segmentation module, we design an Attention Convolutional GRU module (Att-ConvGRU) to learn the long-range context dependence of features. To perceive small or inconspicuous smoke, we design a Multi-scale Context Contrasted Local Feature structure (MCCL) and a Dense Pyramid Pooling Module (DPPM) for improving the representation ability of our network. Extensive experiments validate that our method significantly outperforms existing state-of-art algorithms on smoke datasets, and also obtain satisfactory results on challenging images with inconspicuous smoke and smoke-like objects. [ABSTRACT FROM AUTHOR]

Extreme instance imbalance among categories and combinatorial explosion make the recognition of Human-Object Interaction (HOI) a challenging task. Few studies have addressed both challenges directly. Motivated by the success of few-shot learning that learns a robust model from a few instances, we formulate HOI as a few-shot task in a meta-learning framework to alleviate the above challenges. Due to the fact that the intrinsical characteristic of HOI is diverse and interactive, we propose a Semantic-guided Attentive Prototypes Network (SAPNet) framework to learn a semantic-guided metric space where HOI recognition can be performed by computing distances to attentive prototypes of each class. Specifically, the model generates attentive prototypes guided by the category names of actions and objects, which highlight the commonalities of images from the same class in HOI. In addition, we design two alternative prototypes calculation methods, i.e., Prototypes Shift (PS) approach and Hallucinatory Graph Prototypes (HGP) approach, which explore to learn a suitable category prototypes representations in HOI. Finally, in order to realize the task of few-shot HOI, we reorganize 2 HOI benchmark datasets with 2 split strategies, i.e., HICO-NN, TUHOI-NN, HICO-NF, and TUHOI-NF. Extensive experimental results on these datasets have demonstrated the effectiveness of our proposed SAPNet approach. [ABSTRACT FROM AUTHOR]

Segmentation of the breast ultrasound (BUS) image is an important step for subsequent assessment and diagnosis of breast lesions. Recently, Deep-learning-based methods have achieved satisfactory performance in many computer vision tasks, especially in medical image segmentation. Nevertheless, those methods always require a large number of pixel-wise labeled data that is expensive in medical practices. In this study, we propose a new segmentation method by dense prediction and local fusion of superpixels for breast anatomy with scarce labeled data. First, the proposed method generates superpixels from the BUS image enhanced by histogram equalization, a bilateral filter, and a pyramid mean shift filter. Second, using a convolutional neural network (CNN) and distance metric learning-based classifier, the superpixels are projected onto the embedding space and then classified by calculating the distance between superpixels’ embeddings and the centers of categories. By using superpixels, we can generate a large number of training samples from each BUS image. Therefore, the problem of the scarcity of labeled data can be better solved. To avoid the misclassification of the superpixels, $K$ -nearest neighbor (KNN) is used to reclassify the superpixels within every local region based on the spatial relationships among them. Fivefold cross-validation was taken and the experimental results show that our method outperforms several often used deep-learning methods under the condition of the absence of a large number of labeled data (48 BUS images for training and 12 BUS images for testing). [ABSTRACT FROM AUTHOR]