Synthetic aperture radar (SAR) image classification is a hot topic in the interpretation of SAR images. However, absence effective feature representation and presence speckle noise images make difficult to handle. In order overcome these problems, deep convolutional autoencoder (DCAE) proposed extract features conduct automatically. The network composed eight layers: layer texture features, scale transformation aggregate neighbor information, four layers based on sparse autoencoders optimize...

Deep learning, which represents data by a hierarchical network, has proven to be efficient in computer vision. To investigate the effect of deep features hyperspectral image (HSI) classification, this paper focuses on how extract and utilize HSI classification framework. First, order spectral-spatial information, an improved spatial updated auto-encoder (SDAE), is proposed. SDAE, (DAE), considers sample similarity adding regularization term energy function, updates integrating contextual...

Device-free wireless localization and activity recognition is an emerging technique, which could estimate the location of a person without equipping him/her with any device. It deduces state by analyzing his/her influence on surrounding signals. Therefore, how to characterize human behaviors key question. In this paper, we explore exploit radio image processing approach better Wi-Fi Traditional methods deal channel information (CSI) measurements each independently. However, CSI different...

Abstract Because the reliability of feature for every pixel determines accuracy classification, it is important to design a specialized mining algorithm hyperspectral image classification. We propose learning algorithm, contextual deep learning, which extremely effective On one hand, learning-based extraction can characterize information better than pre-defined algorithm. other spatial Contextual explicitly learns spectral and features via architecture promotes extractor using supervised...

The classification of a synthetic aperture radar (SAR) image is significant yet challenging task, due to the presence speckle noises and absence effective feature representation. Inspired by deep learning technology, novel supervised contractive neural network (DSCNN) for SAR proposed overcome these problems. In order extract spatial features, multiscale patch-based extraction model that consists gray level-gradient co-occurrence matrix, Gabor, histogram oriented gradient descriptors...

Change detection is an important task to identify land-cover changes between the acquisitions at different times. For synthetic aperture radar (SAR) images, inherent speckle noise of images can lead false changed points, which affects change performance. Besides, supervised classifier in framework requires numerous training samples, are generally obtained by manual labeling. In this paper, a novel unsupervised method named saliency-guided deep neural networks (SGDNNs) proposed for SAR image...

The problem of different characters heterogeneous synthetic aperture radar (SAR) images leads to poor performances for transfer learning SAR image classification. To address this issue, a semisupervised model named as deep joint distribution adaptation networks (DJDANs) is proposed from source but similar target image, which aims match the probability distributions between domain and domain. In DJDAN, marginal network developed map features across domains into an augmented common feature...

Ship target detection in synthetic aperture radar (SAR) images is essential for many applications marine monitoring and port security. Though considerable developments have been achieved, there still exist some issues toward multiscale dense ship targets complex inshore scenes. Under such common but challenging situations, it difficult to extract effective information, which drives the missing alarm rate rising dramatically. In scenes, hard disentangle background noise from causes false...

Many emerging applications, such as security surveillance and smart healthcare, have incubated the device-free localization (DFL) technique, which could estimate location of a target without equipping any device. However, current DFL system needs labor-intensive training to learn influence on surrounding wireless signals, so provide model parameters or radio maps for estimation algorithm. To address this issue, we develop novel robust training-free system, named DeFi, directly by refining...

Convolution neural network (CNN) utilizes alternating convolutional and pooling layers to learn representative spatial information when the training samples are sufficient. However, for pixelwise classification of hyperspectral image, some important is neglected by CNN, such as erased operation appearance from lower layers. Moreover, lack a common situation in remote sensing area, which afflicts CNN with overfitting problem. To address aforementioned issues, this paper designs an end-to-end...

Synthetic aperture radar (SAR) image classification is a fundamental process for SAR understanding and interpretation. With the advancement of imaging techniques, it permits to produce higher resolution data extend amount. Therefore, intelligent algorithms high-resolution are demanded. Inspired by deep learning technology, an end-to-end model from original final map developed automatically extract features conduct classification, which named recurrent encoding neural networks (DRENNs). In...

Device-Free simultaneous wireless Localization and Activity Recognition (DFLAR) is a promising novel technique that empowers networks with the ability to perceive location activity of target within its deployment area while not equipping device. This turns traditional into smart context-aware will play an important role in many applications, e.g., city, space, house. Essentially, DFLAR utilizes shadowing effect incurred by on links realize localization recognition. The feature utilized...

Device-free gesture recognition (DFGR) is a promising sensing technique, which can recognize by analyzing its influence on surrounding wireless signals. Most of the DFGR systems are designed based machine learning. However, performance will drop dramatically when testing condition different with training one. Inspired transferrable knowledge learning ability humans, this paper develops practical system metalearning to solve aforementioned problem. Specifically, we design deep network could...

The visual loop-closure detection for autonomous underwater vehicles (AUVs) is a key component to reduce the drift error accumulated in simultaneous localization and mapping tasks. However, due viewpoint changes, textureless images, fast-moving objects, loop closure dramatically changing environments remains challenging problem traditional geometric methods. Inspired by strong feature learning ability of deep neural networks, we propose an method based on variational autoencoder network this...

Hyperspectral target detection (HTD) is an important issue in earth observation, with applications both military and civilian domains. However, conventional representation-based detectors are hindered by the reliance on unknown background dictionary, limited ability to capture nonlinear representations using linear mixing model (LMM), insufficient background-target recognition based handcrafted priors. To address these problems, this paper proposes interpretable representation network that...

The integration of vision-language modalities has been a significant focus in multimodal learning, traditionally relying on Vision-Language Pretrained Models. However, with the advent Large Language Models (LLMs), there notable shift towards incorporating LLMs vision modalities. Following this, training paradigms for into have evolved. Initially, approach was to integrate through pretraining modality integrator, named Single-stage Tuning. It since branched out methods focusing performance...

The classification of polarimetric synthetic aperture radar (PolSAR) image is crucial significance for SAR applications. In this letter, a superpixel restrained deep neural network with multiple decisions (SRDNN-MDs) proposed PolSAR classification, which not only extracts effective spatial features and degrades the influence speckle noises but also deals limited training samples. First, coherency matrix Yamaguchi decomposition are extracted as initial features, segmentation conducted on...

Recent advances in device-free wireless sensing have created the emerging technique of human gesture recognition (DFHGR), which could recognize gestures by analyzing their shadowing effect on surrounding signals. DFHGR has many potential applications fields human-machine interaction, smart home, intelligent space, etc. State-of-the-art work achieved satisfactory accuracy when there are a sufficient number training samples. However, it is time consuming and labor intensive to collect samples,...

The cross-data set knowledge is vital for hyperspectral image classification, which can reduce the dependence on sample quantity by transferring from other data sets and improve training efficiency sharing between different sets. However, due to capturing environment change imaging equipment difference, domain shift troubles exploitation of knowledge. To address aforementioned issue, this article proposes an unsupervised classification method based adversarial adaptation. proposed method,...

Wireless sensing is an emerging technique which empowers wireless devices with additional ability, that is, the ability to sense target location, activity, gesture, vital signs, etc., in a device-free manner by analyzing influence of on surrounding signals. Benefiting from its excellent feature extraction and analysis capability, deep learning has emerged as promising tool realize sensing. However, labor intensive training efforts collecting samples or retraining trained system limit...

Traditional constant false alarm rate (CFAR) based ship target detection methods do not work well in complex conditions, such as multi-scale situations or inshore detection. With the development of deep learning techniques, on convolutional neural networks (CNN) have been applied to solve issues and demonstrated good performance. However, compared with optical datasets, number samples SAR datasets is much smaller, thus limiting Moreover, most state-of-the-art CNN-based detectors that focus...