Feature Maps Of Convolutional Layer Research Articles

Understanding Adversarial Robustness From Feature Maps of Convolutional Layers

Understanding Adversarial Robustness From Feature Maps of Convolutional Layers

Transformed ROIs for capturing visual transformations in videos

Modeling the visual changes that an action brings to a scene is critical for video understanding. Currently, CNNs process one local neighborhood at a time, thus contextual relationships over longer ranges, while still learnable, are indirect. We present TROI, a plug-and-play module for CNNs to reason between mid-level feature representations that are otherwise separated in space and time. The module relates localized visual entities such as hands and interacting objects and transforms their corresponding regions of interest directly in the feature maps of convolutional layers. With TROI, we achieve state-of-the-art action recognition results on the large-scale datasets Something–Something-V2 and EPIC-Kitchens-100.

Vibration-based damage detection for bridges by deep convolutional denoising autoencoder

One of the main challenges for structural damage detection using monitoring data is to acquire features that are sensitive to damages but insensitive to noise (e.g. sensor measurement noise) as well as environmental and operational effects (e.g. temperature effect). Inspired by the capabilities of deep learning methods in representation learning, various deep neural networks have been developed to obtain effective damage features from raw vibration data. However, most of the available deep neural networks are supervised, resulting in practical difficulties owing to the lack of damage labels. This article proposes a damage detection strategy based on an unsupervised deep neural network, referred to as deep convolutional denoising autoencoder, which accepts multi-dimensional cross-correlation functions as input. The strategy aims to extract damage features from field measurements of undamaged structures under the influence of noise and temperature uncertainties. In the proposed strategy, cross-correlation functions of vibration data are first calculated as basic features; then deep convolutional denoising autoencoder is developed to reconstruct cross-correlation functions from their noise-corrupted versions to extract desired features; exponentially weighted moving average control charts are finally established for these features to identify minor structural damages. The strategy is evaluated through a numerical simply supported beam model and an experimental continuous beam model. The mechanism of deep convolutional denoising autoencoder to extract damage features is interpreted by visualizing feature maps of convolutional layers in the encoder. It is found that these layers perform rough estimations of modal properties and preserve the damage information as the general trend of these properties in multiple extra frequency bands. The results show that the proposed strategy is competent for structural damage detection under the exposed environment and worth further exploring its capabilities in applications of real bridges.

SODNet: small object detection using deconvolutional neural network

Convolution neural network (CNN) is an efficient technique to detect objects in various kinds of images, especially for microaneurysm (MA) of diabetic retinopathy in retinal fundus image. This study proposes a deconvolutional neural network to accurately discriminate MA from non-MA. The deconvolution, instead of pooling operation, is embedded into the CNN to recover the erased details of feature maps of convolutional layers. Three types of images are collected for training and predicting. Furthermore, the extracted features are fed into the fully-connected layers to classify using a softmax layer. Experimental results demonstrate that the proposed method can achieve significant sensitivity and accuracy on multiple public datasets, in comparison to the state-of-the-art. For Retinopathy Online Challenge dataset, the sensitivity and accuracy are improved up to 0.798 and 0.986, respectively.

PCAPooL: unsupervised feature learning for face recognition using PCA, LBP, and pyramid pooling

Human face is a widely used biometric modality for verification and revealing the identity of a person. In spite of a great deal of research on face recognition, it still is a challenging issue. Recently, the outstanding performance of deep learning has attracted a good deal of research interest for face recognition. In comparison with hand-engineered features, learning-based face features have proven their superiority in encoding discriminative information. Inspired by deep learning, we introduce a simple and efficient unsupervised feature learning scheme for face recognition. This scheme employs principle component analysis (PCA), local binary pattern (LBP), and pyramid pooling. Following the architecture of a convolutional neural network, the proposed scheme contains three types of layers: convolutional, nonlinear, and pooling layers. PCA is used to learn a filter bank for the convolutional layer. This is followed by LBP operator that encodes the local texture and adds nonlinearity to the feature maps of convolutional layer, which are then pooled using spatial pyramid pooling. To corroborate the effectiveness of the scheme (which we call as PCAPool), extensive experiments were performed on challenging benchmark databases: FERET, Yale, Extended Yale B, AR, and multi-PIE. The comparison reveals that PCAPool performs better than the state-of-the-art methods.

Mining the displacement of max-pooling for text recognition

The max-pooling operation in convolutional neural networks (CNNs) downsamples the feature maps of convolutional layers. However, in doing so, it loses some spatial information. In this paper, we extract a novel feature from pooling layers, called displacement features, and combine them with the features resulting from max-pooling to capture the structural deformations for text recognition tasks. The displacement features record the location of the maximal value in a max-pooling operation. Furthermore, we analyze and mine the class-wise trends of the displacement features. The extensive experimental results and discussions demonstrate that the proposed displacement features can improve the performance of the CNN based architectures and tackle the issues with the structural deformations of max-pooling in the text recognition tasks.

Minutiae-Based Weighting Aggregation of Deep Convolutional Features for Vein Recognition

Deep convolutional neural network (DCNN) has achieved an outstanding performance in large-scale image recognition task because of its discriminative feature representation ability, and pre-trained DCNN models trained for one task have also been applied to domains that are different from their original purposes. Inspired by this idea, a novel hand-dorsa vein recognition model is constructed by adopting DCNN pre-trained on a large-scale database as a universal feature descriptor. However, due to the sparse distribution property of vein information, it is difficult to employ pre-trained DCNN model to extract discriminative deep convolutional features. Therefore, to obtain useful and discriminative deep convolutional features, a novel minutiae-based weighting aggregation (MWA) method is proposed. In specific, the proposed global max-pooling of preserving spatial position information is applied on the feature maps of convolutional layer to localize the minutiae of vein information, and then the minutiae feature of vein image is regarded as the mask that is named as minutiae feature mask, to select deep convolutional features that contain minutiae feature information of vein image. The final feature representation is formed by concatenating each selected deep convolutional feature that is generated by each minutiae feature mask. Series rigorous experiments on the lab-made database are conducted to evidence the effectiveness and feasibility of the proposed MWA for vein recognition. What’s more, an additional experiment with subset of PolyU database illustrates its generalization ability and robustness.