Center Loss Function Research Articles

Classification of jujube defects in small data sets based on transfer learning

Although convolutional neural networks have achieved success in the field of image classification, there are still challenges in the field of agricultural product quality sorting such as machine vision-based jujube defects detection. The performance of jujube defect detection mainly depends on the feature extraction and the classifier used. Due to the diversity of the jujube materials and the variability of the testing environment, the traditional method of manually extracting the features often fails to meet the requirements of practical application. In this paper, a jujube sorting model in small data sets based on convolutional neural network and transfer learning is proposed to meet the actual demand of jujube defects detection. Firstly, the original images collected from the actual jujube sorting production line were pre-processed, and the data were augmented to establish a data set of five categories of jujube defects. The original CNN model is then improved by embedding the SE module and using the triplet loss function and the center loss function to replace the softmax loss function. Finally, the depth pre-training model on the ImageNet image data set was used to conduct training on the jujube defects data set, so that the parameters of the pre-training model could fit the parameter distribution of the jujube defects image, and the parameter distribution was transferred to the jujube defects data set to complete the transfer of the model and realize the detection and classification of the jujube defects. The classification results are visualized by heatmap through the analysis of classification accuracy and confusion matrix compared with the comparison models. The experimental results show that the SE-ResNet50-CL model optimizes the fine-grained classification problem of jujube defect recognition, and the test accuracy reaches 94.15%. The model has good stability and high recognition accuracy in complex environments.

ELF-Net: Enriching local features network for 3D point cloud classification and semantic segmentation

With the popularity of depth sensors and 3D scanners, 3D point cloud has developed rapidly. 3D scene understanding based on deep learning has become a research hotspot. However, many existing networks failed to fully consider the local structures of point clouds, limiting their abilities to exploit the complicated relationships between points. In this paper, we propose Enriching Local Features Network (ELF-Net), which enriches local features of point clouds. We propose Local Points Encoding Module (LPEM) and Feature Concatenate Module (FCM) in our network. Specifically, LPEM is designed to encode the information of eight orientations and 3D coordinate information of local points. We stack the encoding units to achieve multi-scale representation, which is conducive to obtaining robustness and capturing details of the network. In Set Abstraction (SA) module, we apply farthest point sampling (FPS) method to sample the initial points and ball query method is used to group the neighboring points within a radius. FCM is designed to update the representations of local points by applying graph attention mechanism in local regions, which aims to enrich neighboring point feature representations. Finally, our network also proposes a new multivariate loss function, which combines the Center Loss function and Cross Entropy loss function to act on the classification branch. Experimental results show the effectiveness of our proposed network on ModelNet40 (achieves 92.35% accuracy), ScanNet (achieves 85.46% accuracy) and S3DIS (achieves 86.4% accuracy) datasets.

Sensor Signals-Based Early Dementia Detection System Using Travel Pattern Classification

Dementia is becoming more prevalent due to the aging population in which there is deterioration in memory, thinking, behaviour, and the ability to perform everyday activities. A significant challenge in dementia is achieving an accurate and timely diagnosis. If the patient can have proper medical treatment at an early stage, then the dementia growth can be delayed by months to years. Inefficient travel patterns are one of the first indicators of progressive dementia. In this paper, we propose an early dementia detection system using inhabitant travel pattern classification. We use the environmental passive sensor signals for sensing the movement of the inhabitant. The system segments the movements into travel episodes and classifies them using a recurrent neural network. The advantage of using a recurrent neural network is that it directly deals with the raw movement sensory data and does not require any domain-specific knowledge. Finally, the system handles the unbalanced classes of travel patterns by using the focal loss and enhances the discriminative power of the deeply learned features by the center loss function. We conduct several experiments on real-life datasets to verify the accuracy of the system.

CLSTM: Deep Feature-Based Speech Emotion Recognition Using the Hierarchical ConvLSTM Network

Artificial intelligence, deep learning, and machine learning are dominant sources to use in order to make a system smarter. Nowadays, the smart speech emotion recognition (SER) system is a basic necessity and an emerging research area of digital audio signal processing. However, SER plays an important role with many applications that are related to human–computer interactions (HCI). The existing state-of-the-art SER system has a quite low prediction performance, which needs improvement in order to make it feasible for the real-time commercial applications. The key reason for the low accuracy and the poor prediction rate is the scarceness of the data and a model configuration, which is the most challenging task to build a robust machine learning technique. In this paper, we addressed the limitations of the existing SER systems and proposed a unique artificial intelligence (AI) based system structure for the SER that utilizes the hierarchical blocks of the convolutional long short-term memory (ConvLSTM) with sequence learning. We designed four blocks of ConvLSTM, which is called the local features learning block (LFLB), in order to extract the local emotional features in a hierarchical correlation. The ConvLSTM layers are adopted for input-to-state and state-to-state transition in order to extract the spatial cues by utilizing the convolution operations. We placed four LFLBs in order to extract the spatiotemporal cues in the hierarchical correlational form speech signals using the residual learning strategy. Furthermore, we utilized a novel sequence learning strategy in order to extract the global information and adaptively adjust the relevant global feature weights according to the correlation of the input features. Finally, we used the center loss function with the softmax loss in order to produce the probability of the classes. The center loss increases the final classification results and ensures an accurate prediction as well as shows a conspicuous role in the whole proposed SER scheme. We tested the proposed system over two standard, interactive emotional dyadic motion capture (IEMOCAP) and ryerson audio visual database of emotional speech and song (RAVDESS) speech corpora, and obtained a 75% and an 80% recognition rate, respectively.

Deep spatial-temporal networks for flame detection

Every year, fire accidents cause substantial economic losses and casualties. Being able to detect a fire at the early stage is the only way to avoid notable disasters. Although conventional fire alarm systems (CFAs) that depend on heat and flame sensors are used for a fire safety-catch in our society, they cannot be used effectively for large and open spaces due to performance parameters of the sensors. Recently, most of the state-of-the-art methods for fire detection are evolving based on deep learning (DL) technique. However, it is a difficult task to detect fire from visual scenes due to significant irregularities in the color, size, form, texture and flickering frequency of fire. In the present work, we proposed a two-stage cascaded architecture to improve accuracy. In the first stage, we introduced the Spatio-Temporal network, which efficiently and effectively combines both shape and motion flicker based characteristics to obtain heatmaps of fire regions in the input images. By analyzing the heatmaps with a threshold segmentation method, the candidate of the fire region in the input image can be automatically located. Besides, to minimize false-positive due to some object similar to flame, in the second stage, original image and heatmaps of candidate region are fused for improving abilities of classifier to distinguish whether it is a fire or not. Also, the center loss function is adopted to backpropagate fused features to overcome the impact of intraclass heterogeneity on the representation of features. Furthermore, we tested the proposed method on three different datasets, and the results of our experiments reveal that the proposed method has achieved better performance than the other existing state-of-the-art methods.

Rotation-Invariant Feature Learning in VHR Optical Remote Sensing Images via Nested Siamese Structure With Double Center Loss

Rotation-invariant features are of great importance for object detection and image classification in very-high-resolution (VHR) optical remote sensing images. Though multibranch convolutional neural network (mCNN) has been demonstrated to be very effective for rotation-invariant feature learning, how to effectively train such a network is still an open problem. In this article, a nested Siamese structure (NSS) is proposed for training the mCNN to learn effective rotation-invariant features, which consists of an inner Siamese structure to enhance intraclass cohesion and an outer Siamese structure to enlarge interclass margin. Moreover, a double center loss (DCL) function, in which training samples from the same class are mapped closer to each other while those from different classes are mapped far away to each other, is proposed to train the proposed NSS even with a small amount of training samples. Experimental results over three benchmark data sets demonstrate that the proposed NSS trained by DCL is very effective to encounter rotation varieties when learning features for image classification and outperforms several state-of-the-art rotation-invariant feature learning algorithms even when a small amount of training samples are available.

Finger vein recognition based on lightweight CNN combining center loss and dynamic regularization

Finger vein recognition is one of a new biometric recognition technology, which has a wide range of applications in daily life. However, the quality of finger vein images is less than satisfactory because of the disappointing sensor conditions based on infrared light. This problem results in the inaccuracy of finger rein identification. In order to solve this problem and speed up convergence, this paper introduces a new approach of identifying finger veins using the Convolutional Neural Networks (CNNs) with center loss function and dynamic regularization. The proposed method will makes full use of the labels and then ameliorate results. We compare its performance with several popular loss function, such as softmax loss and triplet loss. Experiments were carried out on the datasets of MMCBNU_6000 and FV-USM, whose results show that not only the proposed loss function minimize the error rate and is less time-consuming, but also avoid overfitting.

Vision-based Indoor Localization Algorithm using Improved ResNet

The output of the residual network fluctuates greatly with the change of the weight parameters, which greatly affects the performance of the residual network. For dealing with this problem, an improved residual network is proposed. Based on the classical residual network, batch normalization, adaptive -dropout random deactivation function and a new loss function are added into the proposed model. Batch normalization is applied to avoid vanishing/exploding gradients. -dropout is applied to increase the stability of the model, which we select different dropout method adaptively by adjusting parameter. The new loss function is composed by cross entropy loss function and center loss function to enhance the inter class dispersion and intra class aggregation. The proposed model is applied to the indoor positioning of mobile robot in the factory environment. The experimental results show that the algorithm can achieve high indoor positioning accuracy under the premise of small training dataset. In the real-time positioning experiment, the accuracy can reach 95.37.

T-Center: A Novel Feature Extraction Approach Towards Large-Scale Iris Recognition

For large-scale iris recognition tasks, the determination of classification thresholds remains a challenging task, especially in practical applications where sample space is growing rapidly. Due to the complexity of iris samples, the classification threshold is difficult to determine with the increase of samples. The key issue to solving such threshold determination problems is to obtain iris feature vectors with more obvious discrimination. Therefore, we train deep convolutional neural networks based on a large number of iris samples to extract iris features. More importantly, an optimized center loss function referred to Tight Center (T-Center) Loss is used to solve the problem of insufficient discrimination caused by the traditional Softmax loss function. In order to evaluate the effectiveness of our proposed method, cosine similarity is used to estimate the similarity between the features on the published iris recognition datasets ND-IRIS-0405, CASIA-Thousand and IITD. Our experiment results prove that the T-Center loss can minimize intra-class variance and maximize inter-class variance, which achieve significant performance on the benchmark experiments.

PA-GAN: A Patch-Attention Based Aggregation Network for Face Recognition in Surveillance

Face recognition in unconstraint surveillance is a complicated problem on account of motion blur, expression variations and low resolution. Recent works have demonstrated that patch-attention is strictly more powerful than convolution in recognition models. In this study, we investigate the task of unconstraint surveillance face recognition. First, a Patch-Attention Generative Adversarial Network (PA-GAN) model is devised to aggregate some robust features on behalf of a set of raw surveillance frames, which not only increases the recognition accuracy but also reduces the computational costs of face matching. Second, an improved center loss function combined with abundant unlabeled surveillance faces is utilized to accurately classify the known identities. With the proposed method, the discriminativeness of the face representations is largely enhanced. Finally, the proposed method is verified in two widely used datasets, IJB-A dataset and QMUL-SurvFace dataset to demonstrate the effectiveness. Evaluation of the algorithm performances in comparison with other state-of-the-art methods indicates that the proposed design can achieve competitive accuracy on both the verification and identification protocols.

RETRACTED: Attention-Based Deep Feature Fusion for the Scene Classification of High-Resolution Remote Sensing Images

Scene classification of high-resolution remote sensing images (HRRSI) is one of the most important means of land-cover classification. Deep learning techniques, especially the convolutional neural network (CNN) have been widely applied to the scene classification of HRRSI due to the advancement of graphic processing units (GPU). However, they tend to extract features from the whole images rather than discriminative regions. The visual attention mechanism can force the CNN to focus on discriminative regions, but it may suffer from the influence of intra-class diversity and repeated texture. Motivated by these problems, we propose an attention-based deep feature fusion (ADFF) framework that constitutes three parts, namely attention maps generated by Gradient-weighted Class Activation Mapping (Grad-CAM), a multiplicative fusion of deep features and the center-based cross-entropy loss function. First of all, we propose to make attention maps generated by Grad-CAM as an explicit input in order to force the network to concentrate on discriminative regions. Then, deep features derived from original images and attention maps are proposed to be fused by multiplicative fusion in order to consider both improved abilities to distinguish scenes of repeated texture and the salient regions. Finally, the center-based cross-entropy loss function that utilizes both the cross-entropy loss and center loss function is proposed to backpropagate fused features so as to reduce the effect of intra-class diversity on feature representations. The proposed ADFF architecture is tested on three benchmark datasets to show its performance in scene classification. The experiments confirm that the proposed method outperforms most competitive scene classification methods with an average overall accuracy of 94% under different training ratios.

Alternately Updated Spectral–Spatial Convolution Network for the Classification of Hyperspectral Images

The connection structure in the convolutional layers of most deep learning-based algorithms used for the classification of hyperspectral images (HSIs) has typically been in the forward direction. In this study, an end-to-end alternately updated spectral–spatial convolutional network (AUSSC) with a recurrent feedback structure is used to learn refined spectral and spatial features for HSI classification. The proposed AUSSC includes alternating updated blocks in which each layer serves as both an input and an output for the other layers. The AUSSC can refine spectral and spatial features many times under fixed parameters. A center loss function is introduced as an auxiliary objective function to improve the discrimination of features acquired by the model. Additionally, the AUSSC utilizes smaller convolutional kernels than other convolutional neural network (CNN)-based methods to reduce the number of parameters and alleviate overfitting. The proposed method was implemented on four HSI data sets, as follows: Indian Pines, Kennedy Space Center, Salinas Scene, and Houston. Experimental results demonstrated that the proposed AUSSC outperformed the HSI classification accuracy obtained by state-of-the-art deep learning-based methods with a small number of training samples.

T-Center: A Novel Discriminative Feature Extraction Approach for Iris Recognition

For large-scale iris recognition tasks, the determination of classification thresholds remains a challenging task, especially in practical applications where sample space is growing rapidly. Due to the complexity of iris samples, the classification threshold is difficult to determine with the increase of samples. The key issue to solving such threshold determination problems is to obtain iris feature vectors with more obvious discrimination. Therefore, we train deep convolutional neural networks based on a large number of iris samples to extract iris features. More importantly, an optimized center loss function referred to Tight Center (T -Center) Loss is used to solve the problem of insufficient discrimination caused by Softmax loss function. In order to evaluate the effectiveness of our proposed method, we use cosine similarity to estimate the similarity between the features on the published datasets CASIA-IrisV4 and IITD2.0. Our experiment results demonstrate that the T -Center loss can minimize intra-class variance and maximize inter-class variance, which achieve significant performance on the benchmark experiments.

Deep Center Supervised Hash for Fast Image Retrieval

With the rapid growth of image data, the current mainstream image retrieval method uses the handcraft visual feature coding, which lack of learning ability, resulting in its image expression is not strong, and the large visual feature dimension which seriously restricting its image retrieval performance. The basic idea of this paper is to learn the hash center of each class, so that the distance of intra-class is closer in Hamming space, while maintaining the Hamming distance between the interclass. Firstly, the deep implicit relationship of the training image is extracted by using the robust learning ability of the CNN, and enhance the distinguishing and expressive ability of the image hash feature. In particular, we have designed a CNN architecture in which a hash layer is added to encourage the input image to approximate discrete value (e.g. +1/-1). At the same time, we have carefully designed a hash center loss function, so that the distance between the intra-class to disperse, closer within the intra-class distance, and simultaneously imposing regularization on the real-valued outputs to approximate desired discrete values. We have conducted evolution on CIFAR-10 and NUS-WIDE images benchmarks, demonstrating that our approach can provide superior image search accuracy than other state-of-the-art hashing methods.