Traditional Convolutional Neural Network Models Research Articles

Application of a Hybrid Model Based on a Convolutional Auto-Encoder and Convolutional Neural Network in Object-Oriented Remote Sensing Classification

Variation in the format and classification requirements for remote sensing data makes establishing a standard remote sensing sample dataset difficult. As a result, few remote sensing deep neural network models have been widely accepted. We propose a hybrid deep neural network model based on a convolutional auto-encoder and a complementary convolutional neural network to solve this problem. The convolutional auto-encoder supports feature extraction and data dimension reduction of remote sensing data. The extracted features are input into the convolutional neural network and subsequently classified. Experimental results show that in the proposed model, the classification accuracy increases from 0.916 to 0.944, compared to a traditional convolutional neural network model; furthermore, the number of training runs is reduced from 40,000 to 22,000, and the number of labelled samples can be reduced by more than half, all while ensuring a classification accuracy of no less than 0.9, which suggests the effectiveness and feasibility of the proposed model.

CS-CNN: Enabling Robust and Efficient Convolutional Neural Networks Inference for Internet-of-Things Applications

Recent advances have shown that convolutional neural networks (CNNs) perform excellent in the tasks of image classification and face recognition when the size of data sets is sufficiently large, i.e., over hundreds of thousands training images. Nevertheless, when public data sets are not suitable for training the model for new application scenarios, it is painful to obtain sufficient training examples, especially when the samples have to be labeled manually. Besides, training and inference using CNNs requires significant resources of energy, computation, and memory usage. Therefore, implanting deep CNN models trained and executed on high performance GPU clusters to resource constrained devices, i.e., Internet of Things (IoT) devices, which have permeated into every aspect of modern life, is not appropriate and impractical. Compression technology is an important and popularly used tool to accelerate the training and inference of the CNN models. In this paper, we aim for a step forward in this area: we propose a new compressed CNN model termed CS-CNN for image classification by incorporating the theory of compressive sensing at the input layer of the CNN models to both reduce the resources consumption (evaluated as computation time in this paper) and a required number of training samples. According to our extensive evaluations on the multiple public data sets for deep learning tasks, e.g., MINST and CIFAR-10, using different metrics, we illustrate that the CS-CNN is able to speed up the process of training and inference by a factor of magnitude. Meanwhile, it achieves higher classification accuracy compared with the traditional large CNN models when the size of training database is small.

Deep Learning for Detection of Object-Based Forgery in Advanced Video

Passive video forensics has drawn much attention in recent years. However, research on detection of object-based forgery, especially for forged video encoded with advanced codec frameworks, is still a great challenge. In this paper, we propose a deep learning-based approach to detect object-based forgery in the advanced video. The presented deep learning approach utilizes a convolutional neural network (CNN) to automatically extract high-dimension features from the input image patches. Different from the traditional CNN models used in computer vision domain, we let video frames go through three preprocessing layers before being fed into our CNN model. They include a frame absolute difference layer to cut down temporal redundancy between video frames, a max pooling layer to reduce computational complexity of image convolution, and a high-pass filter layer to enhance the residual signal left by video forgery. In addition, an asymmetric data augmentation strategy has been established to get a similar number of positive and negative image patches before the training. The experiments have demonstrated that the proposed CNN-based model with the preprocessing layers has achieved excellent results.

GCRNN: Group-Constrained Convolutional Recurrent Neural Network.

In this paper, we propose a new end-to-end deep neural network model for time-series classification (TSC) with emphasis on both the accuracy and the interpretation. The proposed model contains a convolutional network component to extract high-level features and a recurrent network component to enhance the modeling of the temporal characteristics of TS data. In addition, a feedforward fully connected network with the sparse group lasso (SGL) regularization is used to generate the final classification. The proposed architecture not only achieves satisfying classification accuracy, but also obtains good interpretability through the SGL regularization. All these networks are connected and jointly trained in an end-to-end framework, and it can be generally applied to TSC tasks across different domains without the efforts of feature engineering. Our experiments in various TS data sets show that the proposed model outperforms the traditional convolutional neural network model for the classification accuracy, and also demonstrate how the SGL contributes to a better model interpretation.

Image Classification Using Biomimetic Pattern Recognition with Convolutional Neural Networks Features.

As a typical deep-learning model, Convolutional Neural Networks (CNNs) can be exploited to automatically extract features from images using the hierarchical structure inspired by mammalian visual system. For image classification tasks, traditional CNN models employ the softmax function for classification. However, owing to the limited capacity of the softmax function, there are some shortcomings of traditional CNN models in image classification. To deal with this problem, a new method combining Biomimetic Pattern Recognition (BPR) with CNNs is proposed for image classification. BPR performs class recognition by a union of geometrical cover sets in a high-dimensional feature space and therefore can overcome some disadvantages of traditional pattern recognition. The proposed method is evaluated on three famous image classification benchmarks, that is, MNIST, AR, and CIFAR-10. The classification accuracies of the proposed method for the three datasets are 99.01%, 98.40%, and 87.11%, respectively, which are much higher in comparison with the other four methods in most cases.

Learning Rotation-Invariant Convolutional Neural Networks for Object Detection in VHR Optical Remote Sensing Images

Object detection in very high resolution optical remote sensing images is a fundamental problem faced for remote sensing image analysis. Due to the advances of powerful feature representations, machine-learning-based object detection is receiving increasing attention. Although numerous feature representations exist, most of them are handcrafted or shallow-learning-based features. As the object detection task becomes more challenging, their description capability becomes limited or even impoverished. More recently, deep learning algorithms, especially convolutional neural networks (CNNs), have shown their much stronger feature representation power in computer vision. Despite the progress made in nature scene images, it is problematic to directly use the CNN feature for object detection in optical remote sensing images because it is difficult to effectively deal with the problem of object rotation variations. To address this problem, this paper proposes a novel and effective approach to learn a rotation-invariant CNN (RICNN) model for advancing the performance of object detection, which is achieved by introducing and learning a new rotation-invariant layer on the basis of the existing CNN architectures. However, different from the training of traditional CNN models that only optimizes the multinomial logistic regression objective, our RICNN model is trained by optimizing a new objective function via imposing a regularization constraint, which explicitly enforces the feature representations of the training samples before and after rotating to be mapped close to each other, hence achieving rotation invariance. To facilitate training, we first train the rotation-invariant layer and then domain-specifically fine-tune the whole RICNN network to further boost the performance. Comprehensive evaluations on a publicly available ten-class object detection data set demonstrate the effectiveness of the proposed method.