Invariant feature extraction for image classification via multi-channel convolutional neural network

Mei, Shaohui; Jiang, Ruoqiao; Ji, Jingyu; Sun, Jian‐Ping; Yang, Peng; Zhang, Yifan

doi:10.1109/ispacs.2017.8266528

Cited by 13 publications

(5 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In other words, the diffusive factor should possess the ability of edge detector in the image with speckle noise. For these reasons, the value of q should be calculated by the formula (4).…”

Section: B Principle Of Speckle Reducing Anisotropic Diffusionmentioning

confidence: 99%

See 1 more Smart Citation

SAR Image Invariant Feature Extraction by Anisotropic Diffusion and Multi-Gray Level Simplified PCNN

Kou

Yue

et al. 2019

IEEE Access

View full text Add to dashboard Cite

Besides the target in synthetic aperture radar (SAR) image, the structural information remained in speckle and shadow is also very important when we extract the invariant feature of SAR image. However, the output of the classical pulse coupled neural network (PCNN) is a binary value. All the information remained in speckle and shadow is lost, when the classical PCNN is used to process SAR image directly. To overcome this problem, a multi-gray level simplified PCNN (MSPCNN) is proposed in this paper. All the useful information in speckle, shadow, and target can be considered when we use MSPCNN to compute the invariant feature of the SAR target. In order to suppress the negative influence of speckle noise and keep the useful structured information, the improved speckle reducing algorithm (SRAD) is used first. By an adaptive threshold delt0, we can keep the remained speckle in different SAR images to an ideal same level. The negative influence of the remained speckle to different targets' signature becomes basically the same and can be ignored. Combining with SRAD and MSPCNN, the SAR image invariant feature extraction scheme AD-MSPCNN is put forward. After analyzing the performance of different signature computing methods, we splice the time signature and the entropy signature together and use the new spliced vector as the invariant feature of the SAR target. The validity and robustness of AD-MSPCNN are proved by the experimental results on MSTAR database. INDEX TERMS Anisotropic diffusion, feature extraction, pulse coupled neural network, speckle reducing, SAR image signature.

show abstract

Section: B Principle Of Speckle Reducing Anisotropic Diffusionmentioning

confidence: 99%

“…The presence of speckle noise affects both the human interpretation of SAR images and the automated feature extraction techniques [1], [2]. Therefore, it may be failed, when the conventional optical image feature extraction algorithms [3], [4] used to process SAR image directly.…”

Section: Introductionmentioning

confidence: 99%

SAR Image Invariant Feature Extraction by Anisotropic Diffusion and Multi-Gray Level Simplified PCNN

Kou

Yue

et al. 2019

IEEE Access

View full text Add to dashboard Cite

show abstract

“…We evaluate the proposed method on two publicly available benchmark datasets: Mnist-rot-12K and NWPU VHR-10, to validate rotationinvariant performance by comparing it with state-of-the-art rotation-invariant algorithms, including RICNN [11], TI-Pooling [12] and traditional CNN features. This work is an extension of our preliminary work published in [15]. Here, we improve the rotation-invariant feature extraction algorithm by designing new loss function, and conduct more extensive and comprehensive experiments including verification analysis and image segmentation to verify the performance of object detection in VHR optical remote sensing images.…”

Section: Introductionmentioning

confidence: 97%

Rotation-Invariant Feature Learning for Object Detection in VHR Optical Remote Sensing Images by Double-Net

et al. 2020

Self Cite

View full text Add to dashboard Cite

Rotation-invariant feature extraction is crucial for object detection in Very High Resolution (VHR) optical remote sensing images. Although convolutional neural networks (CNNs) are good at extracting the translation-invariant features and have been widely applied in computer vision, it is still a challenging problem for CNNs to extract rotation-invariant features in VHR optical remote sensing images. In this paper we present a novel Double-Net with sample pairs from the same class as inputs to improve the performance of object detection and classification in VHR optical remote sensing images. Specifically, the proposed Double-Net contains multiple channels of CNNs in which each channel refers to a specific rotation direction and all CNNs share identical weights. Based on the output features of all channels, multiple instance learning algorithm is employed to extract the final rotation-invariant features. Experimental results on two publicly available benchmark datasets, namely Mnist-rot-12K and NWPU VHR-10, demonstrate that the presented Double-Net outperforms existing approaches on the performance of rotation-invariant feature extraction, which is especially effective under the situation of small training samples. INDEX TERMS Convolution neural network (CNN), feature learning, object detection, rotation-invariant, very high resolution (VHR).

show abstract

“…So, in order to extract image features with more discriminability and form effective image representation, a lot of research has been done on feature extraction algorithms. In recent ten years, it has experienced a development process from extracting shallow layer features based on Scale-invariant feature transform (SIFT) [1], speeded up robust features (SURF) [2] algorithms and embedding coding method in combination with bag of words (BOW) [3,4], fisher vector (FV) [5] and vector of local aggregated descriptors (VLAD) [6] to extracting deep layer features based on the deep convolutional neural network.…”

Section: Introductionmentioning

confidence: 99%

A Novel Feature Aggregation Approach for Image Retrieval Using Local and Global Features

Li¹,

He²,

Ma³

et al. 2022

Computer Modeling in Engineering &Amp; Sciences

View full text Add to dashboard Cite

The current deep convolution features based on retrieval methods cannot fully use the characteristics of the salient image regions. Also, they cannot effectively suppress the background noises, so it is a challenging task to retrieve objects in cluttered scenarios. To solve the problem, we propose a new image retrieval method that employs a novel feature aggregation approach with an attention mechanism and utilizes a combination of local and global features. The method first extracts global and local features of the input image and then selects keypoints from local features by using the attention mechanism. After that, the feature aggregation mechanism aggregates the keypoints to a compact vector representation according to the scores evaluated by the attention mechanism. The core of the aggregation mechanism is to allow features with high scores to participate in residual operations of all cluster centers. Finally, we get the improved image representation by fusing aggregated feature descriptor and global feature of the input image. To effectively evaluate the proposed method, we have carried out a series of experiments on large-scale image datasets and compared them with other state-of-the-art methods. Experiments show that this method greatly improves the precision of image retrieval and computational efficiency.

show abstract

Invariant feature extraction for image classification via multi-channel convolutional neural network

Cited by 13 publications

References 4 publications

SAR Image Invariant Feature Extraction by Anisotropic Diffusion and Multi-Gray Level Simplified PCNN

SAR Image Invariant Feature Extraction by Anisotropic Diffusion and Multi-Gray Level Simplified PCNN

Rotation-Invariant Feature Learning for Object Detection in VHR Optical Remote Sensing Images by Double-Net

A Novel Feature Aggregation Approach for Image Retrieval Using Local and Global Features

Contact Info

Product

Resources

About