Search

In the field of remote sensing, change detection is a crucial study area. Deep learning has made significant strides in the study of remote sensing image change detection during the past few years. Deep learning techniques still have some drawbacks. The global context cannot be modeled by convolutional neural networks due to the receptive field's restrictions. When extracting visual characteristics, the neural network does not concentrate more on the change region, which results in poor distinction between change and no-change regions. To address these problems, we propose networks with large receptive fields (LRFs) and difference image enhancement. First, we design the LRF strategy. It employs a long kernel shape in one spatial dimension for obtaining a long range of relations. Keeping a narrow kernel size in the other spatial dimension can extract local context information while avoiding interference from irrelevant regions. To focus on the changing features, we design the image difference enhancement (IDE) method, which decreases the distance between invariant features and enlarges the distance between changing features. In addition, we design the cross-channel interaction (CNI) strategy, which models the relationship between feature map channels and extracts feature representations through local CNI. On the CDD, WHU-CD, and LEVIR-CD public datasets, we conducted comprehensive experiments. According to the experimental results, our proposed LRDE-Net performs better than other state-of-the-art change detection techniques, and the change regions are more precisely identified. It can better cope with seasonal changes, light intensity, and other pseudochange disturbances.

The existing salient object detection in optical remote sensing images methods mostly employ the same strategy to handle features at different levels without fully considering the distinct characteristics these features possess at various levels. This results in the neglect of some high-level semantics and low-level details during the feature extraction process. Furthermore, existing methods often rely on simple convolution operations to construct modules for feature extraction and fusion. Due to the inherent locality of convolution operations, these models are limited in their performance. To address these challenges, we propose a novel progressive complementation network with semantics and details (SDPCNet) consisting of three parts: Deep semantics aggregation module (DSAM), semantics-guided feature complement module (SFCM), and detail feature enhancement module (DFEM). Specifically, the DSAM is applied on the two highest-level features, guided by the global view with global long-range dependencies and local context generated by transformer and dilated convolution. The DSAM deeply delves the semantic information in high-level features to perceive the object positions and alleviate the adverse effects of cluttered backgrounds. The SFCM operates on the intermediate two levels of features, performing global correlation modeling on the aggregated cross-level features. It enhances multiscale semantic information and edge details using multiple sets of dilated convolutions to address the challenges posed by the uncertainty in the size and number of salient objects. The DFEM acts on the lowest two levels of features, enhancing edge details in spatial dimension and emphasizing semantics in different channel dimensions. It is then fused with high-level features to augment feature diversity and reduce the impact of background noise. Extensive experiments conducted on the ORSSD, EORSSD, and ORSI-4199 datasets demonstrate that our proposed SDPCNet outperforms 23 state-of-the-art methods across eight evaluation metrics.

In recent years, with the development of deep learning, research on end-to-end mispronunciation detection and diagnosis(MDD) methods has been further promoted. At present, research on end-to-end mispronunciation detection and diagnosis is gradually emerging. Most end-to-end mispronunciation detection and diagnosis methods are based on the CNN-RNN-CTC network structure. To improve the performance of end-to-end mispronunciation detection and diagnosis systems, this paper proposes an end-to-end multi-feature and multi-modal mispronunciation detection and diagnosis method based on the Squeezeformer encoder. The model uses Squeezeformer as an audio encoder, a Bi-LSTM network as a phoneme encoder, and Transformer as a decoder. The model fuses phoneme information before speech encoding and decoding, respectively, and uses a secondary decoding mechanism during the decoding process. This study further incorporated phoneme information in the encoding process so that the model could learn the intrinsic characteristics of the speaker’s pronunciation content. The decoding process uses a secondary decoding mechanism to send the sequence decoded by the model to the decoder for decoding again, which solves the problem of no a priori knowledge at the decoder end in the first decoding stage, thus improving the performance of mispronunciation detection and diagnosis. In this study, experiments were conducted on the PSC-Reading Mandarin mispronunciation detection and diagnosis dataset. Compared with the baseline model, the F1 index improved from 0.4060 to 0.7943, and the diagnostic accuracy improved from 83.93% to 88.45%.

RGB-D salient object detection (SOD) usually describes two modes’ classification or regression problem, namely RGB and depth. The existing RGB-D SOD methods use depth hints to increase the detection performance, meanwhile they focus on the quality of little depth maps. In practical application, the interference of various problems in the acquisition process affects the depth map quality, which dramatically reduces the detection effect. In this paper, to minimize interference in depth mapping and emphasize prominent objects in RGB images, we put forward a layered interactive attention network (LIANet). The whole network model adopts the idea of double branch structure to integrate RGB information and Depth information. The network has three parts: feature coding, layered fusion mechanism, and feature decoding. In the feature coding stage, a simple attention module (SAM) is added, which defines the energy function considering the weights of channel and space dimensions. This module enables the network to learn more discriminating neurons without adding parameters. By refining these neurons, the high-level semantic features of images can be fully mined. The layered interactive fusion module (LIFM) is the most critical part of this paper. This module effectively enhances the cross-modal interaction between RGB features and depth features. The RGB-depth-RGB modulation feedback mechanism successfully eliminates interference in the depth map and accurately highlights the features of salient objects. In addition, we also used mixed losses to optimize further and train our model. Finally, a mass of experiments on six standard datasets demonstrated the importance of the method, and a timely detection speed reaches 30 fps on every dataset.

In modern remote sensing image change detection (CD), convolution neural network (CNN), especially U-shaped structure (UNet), has achieved great success due to their powerful discriminative ability. However, UNet-based CNN networks usually have limitations in modeling global dependencies due to the intrinsic locality of convolution operations. Transformer has recently emerged as an alternative architecture for dense prediction tasks due to the global self-attention mechanism. However, due to the limitation of hardware resources, pure Transformer methods generally lack the ability to capture global information at a low level. Based on these existing problems, we propose STransUNet, which combines Transformer and UNet architecture. STransUNet can not only capture shallow detail features at an early stage but also model global context in high-level feature. In addition, we design an efficient feature fusion module named cross-enhanced adaptive fusion (CEAF). Our model mainly consists of three parts: encoder, fusion module, and decoder. The decoder is a CNN-Transformer hybrid structure. CNN extracts multilevel feature information. Transformer encodes tokenized sequence to capture global context. CEAF module cross-enhances and adaptively fuses bitemporal features to enhance feature representation. In the decoding stage, we introduce a cascaded upsampling decoder (CUP). CUP progressively aggregates low-level CNN features and high-level Transformer features to full resolution. On four public CD datasets, our STransUNet achieves better CD results than six state-of-the-art algorithms.

Target tracking has been a research hotspot in computer vision, and the correlation filtered target tracking algorithm has the benefits of low computational complexity and fast speed. Still, the tracking effect is not good when dealing with complicated circumstances. This paper proposes a multi-feature fusion target repositioning tracking algorithm for the target tracking problem in complex environments. First, a multi-feature weighted fusion algorithm is presented. Since each feature has different advantages in different environments, we combine HOG, CN, ULBP, and image edge features and use the weighted coefficient method to adaptively fuse each feature component. Second, to address the target occlusion problem, an occlusion judgment mechanism is introduced, and the target is re-located by fusion weighted filtering. Third, the scale pool is established, and the scale filter is trained by the classification search method. Finally, an adaptive model update strategy is proposed. We conduct comparison experiments with current mainstream algorithms on the publicly available datasets OTB-2015, VOT2018, UAV123, and TColor-128, respectively, and the experimental results show that our proposed algorithm is more robust in complex scenarios.

When taking pictures of electronic screens or objects with high-frequency textures, people often run across colorful rainbow patterns that are known as “moire”, seriously affecting the image quality and subsequent processing. Current methods for removing moire patterns mostly extract multiscale information by downsampling pooling layers, which may inevitably cause information loss. To address this issue, this paper proposes a demoireing method in the wavelet domain. By employing both discrete wavelet transform (DWT) and inverse discrete wavelet transform (IDWT) instead of traditional downsampling and upsampling, this method can effectively increase the network receptive field without information loss. In addition, to further reconstruct more details of moire patterns, this paper proposes an efficient attention fusion module (EAFM). With a combination of efficient channel attention, spatial attention and local residual learning, this module can self-adaptively learn various weights of feature information at different levels and inspire the network to focus more on effective information such as moire details to improve learning and demoireing performance. Extensive experiments based on public datasets have shown that this suggested method can efficiently remove moire patterns and has a good quantitative and qualitative performance.

At present, occlusion and appearance similarity pose severe challenges to person re-identification tasks. Although many robust deep convolutional neural networks alleviate these problems, convolutional layers with limited receptive fields cannot model global semantic information well. In addition, in the person re-identification model, many metric losses ignore or destroy the intra-class structure of the sample, which makes the model difficult to be optimized. Therefore, we design a discriminative Re-identification model with global-local attention and adaptive weighted rank list loss (GLWR). Specifically, our global-local attention (GL-Attention) learns the semantic context in the channel and spatial dimensions. By learning the dependencies between features, GL-Attention integrates global semantic information into local features to extract discriminative features. Unlike rank list loss, our adaptive weighted rank list loss (WRLL) adaptively assigns weights according to the metric distance between the negative sample and the input image, which further improves the performance of the model. Experimental studies on three public datasets (Market-1501, DukeMTMC-ReID and CUHK03) indicate that the performance of our GLWR is significantly superior to many of the latest algorithms.

In the SAR change detection algorithm based on self-supervised learning, speckle noise reduces the difference image (DI) quality. Therefore, the contrast of the DI is low, and its change area is not significant. Moreover, the preclassification algorithm with the poor robustness makes the classification results of the low-quality DI inaccurate. When the wrong labels are sent into the classification network, the accuracy of the final detection results is reduced. First, to improve the quality of the initial DI, we design an adaptive gamma correction algorithm that adjusts the contrast according to the mean value of the initial DI and the variation coefficient β. The contrast of the new DI generated by this algorithm is higher. Furthermore, to suppress the noise, we adopt a new algorithm based on popular ranking to obtain the saliency map of the new DI. Combining the initial DI with this saliency map, a high-quality DI with a low noise level is obtained. After that, we introduce the structure tensor into the fuzzy local information c-means clustering algorithm (FLICM) to classify the DI more accurately. The new clustering algorithm improves the accuracy of preclassification, especially its hierarchical version. Besides, we use the structure tensor to generate the structure maps of the original images. Finally, according to the prior information obtained from the preclassification, we use a convolution wavelet neural network (CWNN) to supervise and train the structure maps of the original images. Experimental results show that the DI generated by us is closer to the ground-truth than other methods. Our preclassification algorithm performs better. Our algorithm shows higher detection accuracy for SAR images with strong noise than some advanced change detection algorithms.

The diversity of network attacks poses severe challenges to intrusion detection systems (IDSs). Traditional attack recognition methods usually adopt mining data associations to identify anomalies, which has the disadvantages of a high false alarm rate (FAR), low recognition accuracy (ACC) and poor generalization ability. To ameliorate the comprehensive capabilities of IDS and strengthen network security, we propose a novel intrusion detection method based on the adaptive synthetic sampling (ADASYN) algorithm and an improved convolutional neural network (CNN). First, we use the ADASYN method to balance the sample distribution, which can effectively prevent the model from being sensitive to large samples and ignore small samples. Second, the improved CNN is based on the split convolution module (SPC-CNN), which can increase the diversity of features and eliminate the impact of interchannel information redundancy on model training. Then, an AS-CNN model mixed with ADASYN and SPC-CNN is used for intrusion detection tasks. Finally, the standard NSL-KDD dataset is selected to test AS-CNN. The simulation illustrates that the accuracy is 4.60% and 2.79% higher than that of the traditional CNN and RNN models, and the detection rate (DR) increased by 11.34% and 10.27%, respectively. Additionally, the FAR decreased by 15.58% and 14.57%, respectively, compared with the two models.

Nowadays, visual object tracking becomes a hotspot and difficulty to achieve a real-time and accurate target tracking, but the Siamese network has solved these difficulties because of its good tracking effect and real-time performance. The location of the target in the previous frame is the template, and the similarity matching is carried out in the search area of the current frame. However, it uses Alexnet network with simple structure and fewer layers to extract features, and just uses a score map to predict the final position of the object. Aiming at these problems, in this paper, we propose the Siamese network of fused response map that use the Alexnet network with fine tuning to extract target features, and weight fusion of score maps to estimate the final position of object. Sufficient experiments on the VOT2015 and OTB100 benchmarks validate that our tracker can improve tracking performance, and perform at 60FPS.

Histopathological image retrieval is a key technology for computer-aided diagnosis. However, patients are reluctant to reveal their privacy in histopathological image retrieval. In order to further improve the effectiveness and safety of histopathological image retrieval, this paper proposes a new histopathological retrieval scheme based on asymmetric residual hash (ARH) and DNA coding techniques. In this paper, we first present a novel ARH for histopathological image retrieval to improve the effectiveness of histopathological search scheme, and then we use the 5-D hyperchaotic system to protect patient privacy. Specifically, the contribution consists of four aspects: 1) A histopathology ciphertext domain search scheme was proposed to improve the performance of computer-aided diagnosis. 2) An asymmetric approach was implemented to process histopathological query points and database points, and a novel asymmetric residual hash algorithm was first proposed to improve the accuracy and speed of histopathological image retrieval. 3) The 5-D hyperchaotic system and DNA coding technique are applied to histopathological image retrieval to protect patient privacy. 4) The loss function is constructed and optimized to learn network parameters and hash codes. The simulation experiment was performed on three datasets (Kimia Path24, Kimia Path960, and Malaria), and the results proved the effectiveness of the ARH algorithm. In addition, our proposed search method can resist common types of attacks during histopathological data transmission. Specifically, the MAP of the ARH is 0.9678 on the KIMIA Path24 with the value of hyperparameter is 125 and the length of hash code is 32. The MAP of the ARH is 0.966 on the KIMIA Path960 with the value of hyperparameter is 225 and the length of hash code is 32. The MAP of the ARH is 0.9482 on the Malaria with the value of hyperparameter is 10 and the length of hash code is 24.

Hashing algorithm has attracted great attention in recent years. In order to improve the query speed and retrieval accuracy, this paper proposes an adaptive and asymmetric residual hash (AASH) algorithm based on residual hash, integrated learning, and asymmetric pairwise loss. The specific description of the AASH algorithm is as follows: 1) the integrated learning model is proposed based on transfer learning and multi-feature fusion strategy to learn the database hash code; 2) the residual hash model is proposed based on ResNet-50 to learn the query image hash code; 3) the asymmetric pairwise loss is proposed and the parameters of the residual hash model is optimized based on the database hash code; 4) the algorithm learns the database hash code and the query image hash code in an asymmetric manner, and integrates the feature learning part and the hash-coded part in one frame. The experimental results on three different datasets fully demonstrate that the proposed AASH method has better performance than most symmetric and asymmetric deep hash algorithms. Specifically, the optimal result of the AASH algorithm is 0.971 on Cifar10 when the hyperparameter is 100 and the hash code length is 32. The optimal result of the AASH algorithm is 0.945 on ceil images when the hyperparameter is 10 and the hash code length is 24. The optimal result of the AASH algorithm is 0.945 on FD-XJ when the hyperparameter is 15 and the hash code length is 32. In addition, the algorithm verifies convergence, time loss, and effectiveness.

Computer-aided diagnosis of retinopathy is a research hotspot in the field of medical image classification. Diabetic macular edema (DME) and age-related macular degeneration (AMD) are two common ocular diseases that can result in partial or complete loss of vision. Optical coherence tomography imaging (OCT) is widely applied to the diagnosis of ocular diseases including DME and AMD. In this paper, an automatic method based on deep learning is proposed to detect AME and AMD lesions, in which two publicly available OCT datasets of retina were adopted and a network model with effective feature of reuse feature was applied to solve the problem of small datasets and enhance the adaptation to the difference of different datasets of the approach. Several network models with effective feature of reusable feature were compared and the transfer learning on networks with pre-trained models was realized. CliqueNet achieves better, classification results compared with other network models with a more than 0.98 accuracy and 0.99 of area under the curve (AUC) value finally.