Deep Hashing Based on Class-Discriminated Neighborhood Embedding

Kang, Jian; Fernández-Beltrán, Rubén; Ye, Zhen; Tong, Xiaohua

doi:10.1109/jstars.2020.3027954

Cited by 14 publications

(4 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The real-valued method uses real-valued features for retrieval. The hashing method incorporates hash code learning into the workflow and employs the learned hash codes for retrieval [19]. The benefit of the hashing method is that the hash code requires less space for storage, and it is more efficient to evaluate image similarity using hash codes.…”

Section: Related Workmentioning

confidence: 99%

Remote Sensing Image Retrieval by Deep Attention Hashing With Distance-Adaptive Ranking

Zhang

Zheng

2023

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

With the joint advancement of numerous related fields of remote sensing, the amount of remote sensing data is growing exponentially. As an essential remote sensing Big Data management technique, content-based remote sensing image retrieval has attracted more and more attention. A novel deep attention hashing with distance-adaptive ranking (DAH) is proposed for remote sensing image retrieval in this article. First, a channel-spatial joint attention mechanism is employed for feature extraction of remote sensing images to make the proposed DAH method focus more on the critical details of the remote sensing images and suppress irrelevant regional responses. Second, a novel balanced pairwise weighted loss function is proposed to enable discrete hash codes to participate in neural network training, which contains pairwise weighted similarity loss, classification loss, and quantization loss. The pairwise weighted similarity loss is designed to decrease the impact of the imbalance of positive and negative sample pairs. The classification loss and quantization loss are added to the loss function to decrease background interference and information loss during the quantization phase, respectively. Finally, a distanceadaptive ranking strategy with category-weighted Hamming distance is presented in the retrieval phase to utilize the category probability information fully. Experiments on benchmark datasets compared with state-of-the-art methods demonstrate the effectiveness of the proposed DAH method.

show abstract

Section: Related Workmentioning

confidence: 99%

Remote Sensing Image Retrieval by Deep Attention Hashing With Distance-Adaptive Ranking

Zhang

Zheng

2023

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Reato et al [29] introduced an unsupervised hash codes learning method for accurate and scalable RSIR. Kang et al [30] proposed a new deep-hashing technique based on the class-discriminated neighborhood embedding, which can properly capture the locality structures among the RS scenes and distinguish images classwisely in the Hamming space. Han et al [31] developed a deep cohesion intensive network, which not only improved the retrieval performance of RS images, but also overcame the imbalance problem of RS images.…”

Section: B Deep Learning-based Rsirmentioning

confidence: 99%

Unsupervised Transformer Balanced Hashing for Multispectral Remote Sensing Image Retrieval

Chen

Wang

Lü

et al. 2023

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

For remote sensing (RS) image retrieval task, hashing technology have been extensively researched in recent works. Unsupervised hashing approaches have attracted much attention in the RS data processing field because label collection takes a lot of time. Most of which fail to consider the interactions among the multichannel information of multispectral RS images and the disparity between the hash-like codes space and the Hamming space, which lead to the poor performance of multispectral RS image retrieval. In this article, we tackle these dilemmas with a novel unsupervised hashing approach, namely Unsupervised Transformer Balanced Hashing (UTBH), to utilize a convolutional variational autoencoder architecture with a novel RS transformer to perform effective hash codes learning. We first integrate a convolutional variational autoencoder architecture with a novel RS transformer, which can guide the interactions among the multichannel information of multispectral RS images. Meanwhile, a new objective function is proposed to preserve discrimination of hash codes in the hashing learning process and reduce the disparity between the hash-like codes space and the Hamming space effectively. Finally, experimental results on two multispectral RS image datasets indicate that UTBH approach achieves superior performance over other unsupervised image retrieval approaches.

show abstract

“…Recently, hashing methods have been introduced to RSIR task [21], [22], [37], [38], [39]. In very early works, dataindependent methods are studied.…”

Section: B Hashing For Rsirmentioning

confidence: 99%

Multiview Inherent Graph Hashing for Large-Scale Remote Sensing Image Retrieval

Sun

Shen

et al. 2021

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Remote sensing image retrieval (RSIR) is one of the most challenging tasks in remote sensing (RS) community. With the volume of RS images increases explosively, conventional exhaustive search is often infeasible in real applications. Recently, hashing has attracted increasing attention for RSIR due to significant advantage in terms of computation and storage. Hashing first generates a set of short compact hash codes to encode RS images, and then applies hash codes for effective RSIR. Multiview hashing usually achieves promising RSIR performance by fusing multiples kinds of RS image features. Conventional multiview hashing simply predefines graph Laplacian in each view, which cannot effectively explore underlying similarity structures among RS images. To address this issue, this article proposes a novel multiview inherent graph hashing (MvIGH) for RSIR. MvIGH captures the latent similarities among RS images, and adaptively learns weights of each view to characterize its contribution. In addition, MvIGH further minimizes the quantization errors. We develop an efficient alternating algorithm to solve the formulated optimization problem. The experiments on three public RS image datasets demonstrate the superiority of the proposed method over the existing multiview hashing methods in RSIR tasks.

show abstract

Deep Hashing Based on Class-Discriminated Neighborhood Embedding

Cited by 14 publications

References 51 publications

Remote Sensing Image Retrieval by Deep Attention Hashing With Distance-Adaptive Ranking

Remote Sensing Image Retrieval by Deep Attention Hashing With Distance-Adaptive Ranking

Unsupervised Transformer Balanced Hashing for Multispectral Remote Sensing Image Retrieval

Multiview Inherent Graph Hashing for Large-Scale Remote Sensing Image Retrieval

Contact Info

Product

Resources

About