Basic characteristics for relating design and base station layout design in land mobile communications are provided through a propagation model for path loss prediction. Owing to the rapid annual increase in traffic data, the number of base stations has increased accordingly. Therefore, propagation models for various scenarios and frequency bands are necessitated. To solve problems optimization and creation methods using the propagation model, a path loss prediction method that merges multiple models in machine learning is proposed herein. The method is discussed based on measurement values from Kitakyushu-shi. In machine learning, the selection of input parameters and suppression of overlearning are important for achieving highly accurate predictions. Therefore, the acquisition of conventional models based on the propagation environment and the use of input parameters of high importance are proposed. The prediction accuracy for Kitakyushu-shi using the proposed method indicates a root mean square error (RMSE) of 3.68dB. In addition, predictions are performed in Narashino-shi to confirm the effectiveness of the method in other urban scenarios. Results confirm the effectiveness of the proposed method for the urban scenario in Narashino-shi, and an RMSE of 4.39dB is obtained for the accuracy.

In recent years, the development of visual tracking is getting better and better, but some methods cannot overcome the problem of low accuracy and success rate of tracking. Although there are some trackers will be more accurate, they will cost more time. In order to solve the problem, we propose a reinforced tracker based on Hierarchical Convolutional Features (HCF for short). HOG, color-naming and grayscale features are used with different weights to supplement the convolution features, which can enhance the tracking robustness. At the same time, we improved the model update strategy to save the time costs. This tracker is called RHCF and the code is published on https://github.com/z15846/RHCF. Experiments on the OTB2013 dataset show that our tracker can validly achieve the promotion of the accuracy and success rate.

We propose an augmented-reality-based method for arranging furniture using natural markers extracted from the edges of the walls of rooms. The proposed method extracts natural markers and estimates the camera parameters from single images of rooms using deep neural networks. Experimental results show that in all the measurements, the superimposition error of the proposed method was lower than that of general marker-based methods that use practical-sized markers.

As the basis of fine breeding management and animal husbandry insurance, individual recognition of dairy cattle is an important issue in the animal husbandry management field. Due to the limitations of the traditional method of cow identification, such as being easy to drop and falsify, it can no longer meet the needs of modern intelligent pasture management. In recent years, with the rise of computer vision technology, deep learning has developed rapidly in the field of face recognition. The recognition accuracy has surpassed the level of human face recognition and has been widely used in the production environment. However, research on the facial recognition of large livestock, such as dairy cattle, needs to be developed and improved. According to the idea of a residual network, an improved convolutional neural network (Res_5_2Net) method for individual dairy cow recognition is proposed based on dairy cow facial images in this letter. The recognition accuracy on our self-built cow face database (3012 training sets, 1536 test sets) can reach 94.53%. The experimental results show that the efficiency of identification of dairy cows is effectively improved.

Convolutional Neural Network (CNN) has made extraordinary progress in image classification tasks. However, it is less effective to use CNN directly to detect image manipulation. To address this problem, we propose an image filtering layer and a multi-scale feature fusion module which can guide the model more accurately and effectively to perform image manipulation detection. Through a series of experiments, it is shown that our model achieves improvements on image manipulation detection compared with the previous researches.

Although deep neural networks (DNNs) have achieved high performance across a variety of applications, they can often be deceived by adversarial examples that are generated by adding small perturbations to the original images. Adversaries may generate adversarial examples using the property of transferability, in which adversarial examples that deceive one model can also deceive other models because adversaries do not obtain any information on the DNNs deployed in real scenarios. Recent studies show that adversarial examples with feature space perturbations are more transferable than others. Adversarial training is an effective method to defend against adversarial attacks. However, it results in a decrease in the classification accuracy for natural images, and it is not sufficiently robust against transferable adversarial examples because it does not consider adversarial examples with feature space perturbations. We propose a novel adversarial training method to train DNNs to be robust against transferable adversarial examples and maximize their classification accuracy for natural images. The proposed method trains DNNs to correctly classify natural images and adversarial examples and also minimize the feature differences between them. The robustness of the proposed method was similar to those of the previous adversarial training methods for MNIST dataset and was up to average 6.13% and 9.24% more robust against transfer adversarial examples for CIFAR-10 and CIFAR-100 datasets, respectively. In addition, the proposed method yielded an average classification accuracy that was approximately 0.53%, 6.82%, and 10.60% greater than some state-of-the-art adversarial training methods for all datasets, respectively. The proposed method is robust against a variety of transferable adversarial examples, which enables its implementation in security applications that may benefit from high-performance classification but are at high risk of attack.

This paper presents a physical layer wireless device identification method that uses a convolutional neural network (CNN) operating on a quadrant IQ transition image. This work introduces classification and detection tasks in one process. The proposed method can identify IoT wireless devices by exploiting their RF fingerprints, a technology to identify wireless devices by using unique variations in analog signals. We propose a quadrant IQ image technique to reduce the size of CNN while maintaining accuracy. The CNN utilizes the IQ transition image, which image processing cut out into four-part. An over-the-air experiment is performed on six Zigbee wireless devices to confirm the proposed identification method's validity. The measurement results demonstrate that the proposed method can achieve 99% accuracy with the light-weight CNN model with 36,500 weight parameters in serial use and 146,000 in parallel use. Furthermore, the proposed threshold algorithm can verify the authenticity using one classifier and achieved 80% accuracy for further secured wireless communication. This work also introduces the identification of expanded signals with SNR between 10 to 30dB. As a result, at SNR values above 20dB, the proposals achieve classification and detection accuracies of 87% and 80%, respectively.

Cow behavior monitoring is critical for understanding the current state of cow welfare and developing an effective planning strategy for pasture management, such as early detection of disease and estrus. One of the most powerful and cost-effective methods is a neural-network-based monitoring system that analyzes time series data from inertial sensors attached to cows. For this method, a significant challenge is to improve the quality and quantity of teaching data in the development of neural network models, which requires us to collect data that can cover various realistic conditions and assign labels to them. As a result, the cost of data collection is significantly high. This work proposes a data augmentation method to solve two major quality problems in the collection process of teaching data. One is the difficulty and randomicity of teaching data acquisition and the other is the sensor position changes during actual operation. The proposed method can computationally emulate different rotating states of the collar-type sensor device from the measured acceleration data. Furthermore, it generates data for actions that occur less frequently. The verification results showed significantly higher estimation performance with an average accuracy of over 98% for five main behaviors (feeding, walking, drinking, rumination, and resting) based on learning with long short-term memory (LSTM) network. Compared with the estimation performance without data augmentation, which was insufficient with a minimum of 60.48%, the recognition rate was improved by 2.52-37.05pt for various behaviors. In addition, comparison of different rotation intervals was investigated and a 30-degree increment was selected based on the accuracy performances analysis. In conclusion, the proposed data expansion method can improve the accuracy in cow behavior estimation by a neural network model. Moreover, it contributes to a significant reduction of the teaching data collection cost for machine learning and opens many opportunities for new research.

A method to generate color palettes from images is proposed. Here, deep neural networks (DNN) are utilized in order to consider human perception. Two aspects of human perception are considered; one is attention to image, and the other is human preference for colors. This method first extracts N regions with dominant color categories from the image considering human attention. Here, N is the number of colors in a color palette. Then, the representative color is obtained from each region considering the human preference for color. Two deep neural-net systems are adopted here, one is for estimating the image area which attracts human attention, and the other is for estimating human preferable colors from image regions to obtain representative colors. The former is trained with target images obtained by an eye tracker, and the latter is trained with dataset of color selection by human. Objective and subjective evaluation is performed to show high performance of the proposed system compared with conventional methods.

In recent years, compared with the traditional face super-resolution (SR) algorithm, the face SR based on deep neural network has shown strong performance. Among these methods, attention mechanism has been widely used in face SR because of its strong feature expression ability. However, the existing attention-based face SR methods can not fully mine the missing pixel information of low-resolution (LR) face images (structural prior). And they only consider a single attention mechanism to take advantage of the structure of the face. The use of multi-attention could help to enhance feature representation. In order to solve this problem, we first propose a new pixel attention mechanism, which can recover the structural details of lost pixels. Then, we design an attention fusion module to better integrate the different characteristics of triple attention. Experimental results on FFHQ data sets show that this method is superior to the existing face SR methods based on deep neural network.

This paper addresses single image super-resolution (SR) based on convolutional neural networks (CNNs). It is known that recovery of high-frequency components in output SR images of CNNs learned by the least square errors or least absolute errors is insufficient. To generate realistic high-frequency components, SR methods using generative adversarial networks (GANs), composed of one generator and one discriminator, are developed. However, when the generator tries to induce the discriminator's misjudgment, not only realistic high-frequency components but also some artifacts are generated, and objective indices such as PSNR decrease. To reduce the artifacts in the GAN-based SR methods, we consider the set of all SR images whose square errors between downscaling results and the input image are within a certain range, and propose to apply the metric projection onto this consistent set in the output layers of the generators. The proposed technique guarantees the consistency between output SR images and input images, and the generators with the proposed projection can generate high-frequency components with few artifacts while keeping low-frequency ones as appropriate for the known noise level. Numerical experiments show that the proposed technique reduces artifacts included in the original SR images of a GAN-based SR method while generating realistic high-frequency components with better PSNR values in both noise-free and noisy situations. Since the proposed technique can be integrated into various generators if the downscaling process is known, we can give the consistency to existing methods with the input images without degrading other SR performance.

Machine reading comprehension with multi-hop reasoning always suffers from reasoning path breaking due to the lack of world knowledge, which always results in wrong answer detection. In this paper, we analyze what knowledge the previous work lacks, e.g., dependency relations and commonsense. Based on our analysis, we propose a Multi-dimensional Knowledge enhanced Graph Network, named MKGN, which exploits specific knowledge to repair the knowledge gap in reasoning process. Specifically, our approach incorporates not only entities and dependency relations through various graph neural networks, but also commonsense knowledge by a bidirectional attention mechanism, which aims to enhance representations of both question and contexts. Besides, to make the most of multi-dimensional knowledge, we investigate two kinds of fusion architectures, i.e., in the sequential and parallel manner. Experimental results on HotpotQA dataset demonstrate the effectiveness of our approach and verify that using multi-dimensional knowledge, especially dependency relations and commonsense, can indeed improve the reasoning process and contribute to correct answer detection.

In the user identification (UI) scheme for a multiple-access fading channel based on a randomly generated (0, 1, -1)-signature code, previous studies used the signature code over a noisy multiple-access adder channel, and only the user state information (USI) was decoded by the signature decoder. However, by considering the communication model as a compressed sensing process, it is possible to estimate the channel coefficients while identifying users. In this study, to improve the efficiency of the decoding process, we propose an iterative deep neural network (DNN)-based decoder. Simulation results show that for the randomly generated (0, 1, -1)-signature code, the proposed DNN-based decoder requires less computing time than the classical signal recovery algorithm used in compressed sensing while achieving higher UI and channel estimation (CE) accuracies.

Online feeds are streamed continuously in batches with varied polarities at varying times. The system handling the online feeds must be trained to classify all the varying polarities occurring dynamically. The polarity classification system designed for the online feeds must address two significant challenges: i) stability-plasticity, ii) category-proliferation. The challenges faced in the polarity classification of online feeds can be addressed using the technique of incremental learning, which serves to learn new classes dynamically and also retains the previously learned knowledge. This paper proposes a new incremental learning methodology, ILOF (Incremental Learning of Online Feeds) to classify the feeds by adopting Deep Learning Techniques such as RNN (Recurrent Neural Networks) and LSTM (Long Short Term Memory) and also ELM (Extreme Learning Machine) for addressing the above stated problems. The proposed method creates a separate model for each batch using ELM and incrementally learns from the trained batches. The training of each batch avoids the retraining of old feeds, thus saving training time and memory space. The trained feeds can be discarded when new batch of feeds arrives. Experiments are carried out using the standard datasets comprising of long feeds (IMDB, Sentiment140) and short feeds (Twitter, WhatsApp, and Twitter airline sentiment) and the proposed method showed positive results in terms of better performance and accuracy.

Low power consumption is important in edge artificial intelligence (AI) chips, where power supply is limited. Therefore, we propose reconfigurable neural network accelerator (ReNA), an AI chip that can process both a convolutional layer and fully connected layer with the same structure by reconfiguring the circuit. In addition, we developed tools for pre-evaluation of the performance when a deep neural network (DNN) model is implemented on ReNA. With this approach, we established the flow for the implementation of DNN models on ReNA and evaluated its power consumption. ReNA achieved 1.51TOPS/W in the convolutional layer and 1.38TOPS/W overall in a VGG16 model with a 70% pruning rate.

A backdoor attack is a type of attack method inducing deep neural network (DNN) misclassification. An adversary mixes poison data, which consist of images tampered with adversarial marks at specific locations and of adversarial target classes, into a training dataset. The backdoor model classifies only images with adversarial marks into an adversarial target class and other images into the correct classes. However, the attack performance degrades sharply when the location of the adversarial marks is slightly shifted. An adversarial mark that induces the misclassification of a DNN is usually applied when a picture is taken, so the backdoor attack will have difficulty succeeding in the physical world because the adversarial mark position fluctuates. This paper proposes a new approach in which an adversarial mark is applied using fault injection on the mobile industry processor interface (MIPI) between an image sensor and the image recognition processor. Two independent attack drivers are electrically connected to the MIPI data lane in our attack system. While almost all image signals are transferred from the sensor to the processor without tampering by canceling the attack signal between the two drivers, the adversarial mark is injected into a given location of the image signal by activating the attack signal generated by the two attack drivers. In an experiment, the DNN was implemented on a Raspberry pi 4 to classify MNIST handwritten images transferred from the image sensor over the MIPI. The adversarial mark successfully appeared in a specific small part of the MNIST images using our attack system. The success rate of the backdoor attack using this adversarial mark was 91%, which is much higher than the 18% rate achieved using conventional input image tampering.

Deep learning applications have often been processed in the cloud or on servers. Still, for applications that require privacy protection and real-time processing, the execution environment is moved to edge devices. Edge devices that implement a neural network (NN) are physically accessible to an attacker. Therefore, physical attacks are a risk. Fault attacks on these devices are capable of misleading classification results and can lead to serious accidents. Therefore, we focus on the softmax function and evaluate a fault attack using a clock glitch against NN implemented in an 8-bit microcontroller. The clock glitch is used for fault injection, and the injection timing is controlled by monitoring the power waveform. The specific waveform is enrolled in advance, and the glitch timing pulse is generated by the sum of absolute difference (SAD) matching algorithm. Misclassification can be achieved by appropriately injecting glitches triggered by pattern detection. We propose a countermeasure against fault injection attacks that utilizes the randomization of power waveforms. The SAD matching is disabled by random number initialization on the summation register of the softmax function.

There have been some researchers that investigate the accuracy of the approximation to a function that shows a generating pattern of data by a deep neural network. However, they have confirmed only whether at least one function close to the function showing a generating pattern exists in function classes of deep neural networks whose parameter values are changing. Therefore, we propose a new criterion to infer the approximation accuracy. Our new criterion shows the existence ratio of functions close to the function showing a generating pattern in the function classes. Moreover, we show a deep neural network with a larger number of layers approximates the function showing a generating pattern more accurately than one with a smaller number of layers under the proposed criterion, with numerical simulations.

Many cloud service providers employ specialized hardware accelerators, called neural processing units (NPUs), to accelerate deep neural networks (DNNs). An NPU scheduler is responsible for scheduling incoming user requests and required to satisfy the two, often conflicting, optimization goals: maximizing system throughput and satisfying quality-of-service (QoS) constraints (e.g., deadlines) of individual requests. We propose Layerweaver+, a low-cost layer-wise DNN scheduler for NPUs, which provides both high system throughput and minimal QoS violations. For a serving scenario based on the industry-standard MLPerf inference benchmark, Layerweaver+ significantly improves the system throughput by up to 266.7% over the baseline scheduler serving one DNN at a time.

Object contour detection is a task of extracting the shape created by the boundaries between objects in an image. Conventional methods limit the detection targets to specific categories, or miss-detect edges of patterns inside an object. We propose a new method to represent a contour image where the pixel value is the distance to the boundary. Contour detection becomes a regression problem that estimates this contour image. A deep convolutional network for contour estimation is combined with stereo vision to detect unspecified object contours. Furthermore, thanks to similar inference targets and common network structure, we propose a network that simultaneously estimates both contour and disparity with fully shared weights. As a result of experiments, the multi-tasking network drew a good precision-recall curve, and F-measure was about 0.833 for FlyingThings3D dataset. L1 loss of disparity estimation for the dataset was 2.571. This network reduces the amount of calculation and memory capacity by half, and accuracy drop compared to the dedicated networks is slight. Then we quantize both weights and activations of the network to 3-bit. We devise a dedicated hardware architecture for the quantized CNN and implement it on an FPGA. This circuit uses only internal memory to perform forward propagation calculations, that eliminates high-power external memory accesses. This circuit is a stall-free pixel-by-pixel pipeline, and performs 8 rows, 16 input channels, 16 output channels, 3 by 3 pixels convolution calculations in parallel. The convolution calculation performance at the operating frequency of 250 MHz is 9 TOPs/s.