Search

Aiming at the nonlinear and non‐stationary characteristics of the power load, in order to improve the accuracy of load forecasting, a short‐term load forecasting method based on variational modal decomposition and the combination of convolutional neural network and bi‐directional long short‐term memory (Bi‐LSTM) network is proposed. The load sequence is decomposed into components with different frequencies by variational modal decomposition, and each component is predicted by Bi‐LSTM. Then the convolution neural network is introduced to process multi‐source data. The load, temperature, day and other data are constructed according to the time sliding window as the input, and the convolution neural network is used to extract the effective feature vector, the eigenvector is constructed in a time series and used as the input data of Bi‐LSTM network. Taking the public data set provided by a US public utility department as an example, the proposed model is compared with other methods. The results show that the proposed method can better track the change of load and effectively improve the accuracy of short‐term load forecasting. © 2023 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Accurately‐labeled data is crucial for the training of machine learning models. For singing‐related tasks in the music information retrieval field, accurately‐labeled data is limited because annotating singing is time‐consuming. Several studies create vocal datasets using a two‐step annotation method which creates coarse labels first and then executes a manual calibration procedure. However, manually calibrating coarsely‐labeled singing data is expensive and time‐consuming. To address this problem, in this study we propose a singing‐label calibration framework, which aims to automatically calibrate the coarsely‐labeled singing data with higher accuracy. This framework contains a data augmentation method to generate training and testing data, a reasonable data preprocessing method to handle music audio and symbolic labels, a fully‐convolutional neural network to estimate the difference between coarse labels and accurate labels, and a novel calibration function to correct the coarse labels. Various experiments are conducted to examine the effect of our research. The results show that our model can highly reduce the cost time and slightly increase the labeling accuracy of the manual calibration process. © 2023 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Detecting damages on the body of the saury is important for packaging in the fisheries industry. Moreover, the appearance of a fish is important in fisheries. Fish damage detection relies on manual assembly lines, which are labor‐intensive and time‐consuming. This paper proposes a novel saury damage detection method that automatically localizes the saury body region, determines the head‐tail orientation, and detects damaged parts from the body region. We collected a saury image dataset for training and built a simple naive convolutional neural network (CNN) model without transfer learning for damage discrimination. Seven evaluation metrics, such as accuracy, F1‐score, Area under the ROC curve (AUC), and prediction time, were comprehensively considered to select the optimal parameter configuration including the number of layers, number of filters, filter size, and input image size. We also utilized the transfer learning of eight pretrained CNN models to classify the saury into two categories for damage discrimination. Inception‐v3 and ResNet‐101 achieved the highest accuracy of 98.2% for damage discrimination. Through extensive experiments and discussion of the results, this paper presents clear guidance for choosing pretrained networks for transfer learning in damage detection tasks. © 2023 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

A method for detecting atrial fibrillation (AF) from electrocardiogram (ECG) measured by a 24‐hour Holter electrocardiograph (Holter‐ECG) is proposed using convolutional neural network (CNN). In the preprocessing stage, artifacts and noises on Holter‐ECG are removed by a bandpass filter. The detection method consists of two stages: extraction of abnormal waveforms using one‐dimensional CNN trained with segmented ECG waveform and its spectral entropy, and identification of AF using two‐dimensional CNN trained with segmented ECG spectrogram. A total of 47 520 datasets obtained from Holter‐ECG were prepared, and used for training at both CNN stages. Newly prepared (untrained) datasets of 24‐hour Holter‐ECG of 10 subjects and MIT‐BIH databases are tested, and the proposed method showed sufficient performance for detecting AF, with the accuracy of approximately 90%. This result indicates the feasibility of the proposed method. © 2023 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Since deep low‐frequency tremors are considered to be associated with large earthquakes that occur adjacently on the same subducting plate interface, it is important to investigate tremors that occurred before the establishment of modern seismograph networks such as the High Sensitivity Seismograph Network (Hi‐net). We propose a deep‐learning solution to detect evidence of tremors in scanned images of paper seismogram records from over 50 years ago. In this study, we fine‐tuned a convolutional neural network (CNN) based on the Residual Network, which was pre‐trained using images of synthetic waveforms from our previous study, using a data set comprised of images generated from real seismic data recorded digitally by Hi‐net to facilitate a supervised analysis. The fine‐tuned CNN was able to predict the presence or absence of tremors in the Hi‐net images with an accuracy of 98.64%. Gradient‐weighted Class Activation Mapping heatmaps created to visualize model predictions indicated that the CNN's ability to detect tremors is not degraded by the presence of teleseisms. Once validated using the Hi‐net images, the CNN was applied to paper seismograms recorded from 1966 to 1977 at the Kumano observatory in southwest Japan, operated by Earthquake Research Institute, The University of Tokyo. The CNN showed potential for detecting tremors in scanned images of paper seismogram records from the past, facilitating downstream tasks such as the creation of new tremor catalogs. However, further training using an augmented data set to control for variables such as inconsistent plotting pen thickness is required to develop a universally applicable model. Plain Language Summary: In 2002, deep low‐frequency tremors, which periodically occur with much smaller amplitudes and longer durations than common earthquakes, were discovered owing to the establishment of dense seismic arrays in Japan. These tremors are considered to be associated with megathrust earthquakes because they occur in deeper regions of plate boundaries compared with common earthquakes. Investigating tremors that occurred before the establishment of dense seismic arrays is important since megathrust earthquakes recur on intervals of 100–200 years, and digital records from modern seismic arrays cover only a fraction of that time. In this study, we developed a convolutional neural network based on the Residual Network (ResNet) that extracts tremors from images of waveforms recorded more than 50 years ago, when seismometers drew waveforms directly on drum‐rolled paper using a pen. The proposed ResNet, which was trained on seismogram images generated from synthetic waveforms and real data recorded digitally by modern seismic arrays, was applied to scanned images of paper seismograms recorded from 1966 to 1977 at an observatory in southwest Japan. A list of tremors was successfully obtained; however, further training using data that accounts for variables such as inconsistent plotting pen thickness is required to develop a universally applicable model. Key Points: Convolutional neural network model for detection of deep low‐frequency tremors from seismogram images is proposedThe model trained with seismogram images converted from real seismic data successfully detects tremorsThe detection performances of the trained model for the paper records at the Kumano observatory in southwest Japan are discussed [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel multi‐objective topology optimization with response surface approximation interpolating various topologies, which is derived by deep learning with training data in the actual design space. The response surface is constructed by means of autoencoder with the training data of geometrically diverse shapes of targeted devices, which can express the interpolations that combine the multiple characteristic structures. The global topology search is performed based on the gradient information of the response surface, by which we can obtain the initial shapes for level set optimization to efficiently obtain easily manufactured and excellent Pareto optimal solutions. The effectiveness of the proposed method is demonstrated by optimizing multi‐flux barriers in a synchronous reluctance motor for the objective functions of average torque and torque ripple. © 2022 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Curb detection with a monocular camera is important to assist driving by detecting curbs to avoid accidents. However, various road scenes and curb shapes make it difficult to detect curbs based on a single image. In this study, a novel image‐based curb detection method that is highly efficient, inexpensive, and less complex than LiDAR‐based solutions was developed. A deep learning framework was used to detect curbs from road images automatically. A custom convolutional neural network (CNN) model was built, and another seven pretrained models were fine‐tuned for curb patch classification. The evaluation metrics, such as accuracy, F1‐score, area under curve, and prediction time, were considered comprehensively to select the optimal CNN architecture. Three promising CNN architectures were employed as classification networks and embedded into the YOLO‐v2 framework to construct curb detectors. The detection performance was evaluated in terms of the average precision on an urban road image dataset. By configuring parameters for the optimal CNN architecture, the best detector achieved an average precision of 99.16%, which verifies the effectiveness of the proposed method. © 2022 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Deep residual network (ResNet), one of the mainstream deep learning models, has achieved groundbreaking results in various fields. However, all neurons used in ResNet are based on the McCulloch‐Pitts model which has long been criticized for its oversimplified structure. Accordingly, this paper for the first time proposes a novel dendritic residual network by considering the powerful information processing capacity of dendrites in neurons. Experimental results based on the challenging COVID‐19 prediction problem show the superiority of the proposed method in comparison with other state‐of‐the‐art ones. © 2022 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

A spherical image has a full field of view. Using spherical images can estimate camera pose more accurately than conventional perspective images. It is known that a sphere can be quasi‐uniform sampled by dividing a geodesic dome repeatedly; it results in a representation with rotational invariance approximately. Based on this fact, in this paper we proposed a method of estimating a spherical camera pose based on the geodesic‐dome‐division‐based discrete spherical image representation by modifying the conventional perspective‐image‐oriented convolutional neural network, called PoseNet. First, a spherical image is represented as a geodesic‐dome‐division‐based discrete spherical image. Then, to cope with the six‐connectivity of the pixels of a geodesic‐dome‐division‐based discrete spherical image using the conventional convolutional neural network framework, the discrete spherical image is flattened as a connected five parallelograms and resampled as an eight‐connectivity image. Consequently, via the comparative experiments, it is found that the geodesic‐dome‐division‐based discrete spherical image representation approach can achieve a better result of camera pose estimation than equirectangular image and CubeMap representations. © 2021 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

For Automatic High Beam (AHB), the authors are working on a study to classify beam of car headlights as 'High' and 'Low' by realizing the presence of vehicle on road from nighttime in‐vehicle camera image using deep learning. In general, weights of deep learning model are pre‐trained on a large dataset which are then used as initial weights for other tasks such as image recognition. However, it has been reported that even if the pre‐trained weights are used as the initial weights, it is not a factor of accuracy improvement when domain differs. It is important to pre‐train a model on a dataset that is suitable for target domain. Therefore, we propose a method to generate a pre‐training dataset that can be easily created by using visual explanation which represents where a deep learning model is looking at when performs a task. Then, we applied the proposed pre‐training dataset on the headlight beam classification in the nighttime in‐vehicle camera image and verified the effectiveness of our method. © 2021 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

This paper proposes a new fish bone detection technique using a convolutional neural network (CNN) and synthetic image generation. Semantic segmentation CNNs with supervised learning generally require a large number of training images and their pixel‐wise teaching signals. In fish bone detection, there are two problems with using semantic segmentation CNNs. One is the manual annotations of fish bones and the other is the difficulty of sampling all variations of fish bones with various lengths, angles, and thicknesses. The proposed method, however, generates them by drawing virtual fish bones on X‐ray images. This technique is very useful for reducing the cost of collecting and annotating a dataset. Experimental results have shown that the average F‐measure for the proposed method is 0.747, while that for a normal training method is 0.493. In the proposed method, the CNN successfully detected actual fish bones despite its training only with virtual fish bones. © 2021 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Low-frequency tremors have been widely detected in many tectonic zones, and are often located adjacent to megathrust zones, indicating that their spatiotemporal evolution provides important insights into megathrust events. The envelope correlation method (ECM) is commonly used to detect tremors. However, the ECM also detects regular earthquakes, which requires the separation of these two signals after the initial detection. In addition, signals of tremors are weak, so classifying tremors from noises is also an essential problem. We develop a convolutional neural network (CNN)-based method using a single S-net station located off Sanriku region, Northeast Japan, to classify local earthquakes, tremors, and noise. Along the Japan Trench, especially in a region focused in this study, local earthquakes and tremors occurred in coexistence within a small region, so detection, location, and discrimination of these events are the key to understand the relationship between slow and regular earthquakes. The spectrograms of the three-component velocity waveforms that were recorded during 16 August 2016 to 14 August 2018 are used as the training and test datasets for the CNN. The CNN successfully classified 100%, 96%, and 98% of the earthquakes, tremors, and noise, respectively. We also showed a successful application of our method to continuous waveform data including a tremor to explore the feasibility of the proposed method in classifying tremors and noise in continuous streaming data. The output probabilities for the true classifications decrease with increasing epicentral distance and/or decreasing event magnitude. This highlights the need to train the CNN using tremors proximal to the seismic stations for detecting tremors using multiple stations. [ABSTRACT FROM AUTHOR]

More and more factories and commercial buildings are installing combined heat and power (CHP) systems that include various energy storage devices. To reduce the energy cost of CHPs, optimal operation plans to satisfy time‐varying energy demands with minimum energy cost are required. Conventional operation planning methods using optimized calculation have an issue with long computing time. To address this, we adapted a convolutional neural network (CNN) to learn the time‐series features of input–output pairs of optimized calculations. The architecture of the proposed CNN has the following key features: (i) the application of one‐dimensional convolutional filters to extract the time‐series features of input data and (ii) the elimination of pooling layers that compress the input data. We also designed rule‐based flows to remove constraint violations from the generated operation plans. Our evaluations showed that the proposed method could generate operation plans with the average error of 3.3%, which is much lower than the 8.0% average error when using the conventional CNN with pooling calculation. © 2021 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Most face detection and recognition tasks are based on the training of intact facial images and corresponding labels. Both the three‐dimensional (3D) structure and two‐dimensional (2D) appearance from the frontal view of human faces are bilaterally symmetrical in general. However, sometimes, illumination on the left and right halves of faces is uneven. In such cases, the symmetrical characteristic of human faces can facilitate expressing distinct identity information. This is because even if one side of the facial image is corrupted by noise, the opposite side can still be used for feature extraction. This paper proposes an automatic selection of the better half of the face using only a half‐face for identity recognition. Unlike the MegaFace challenge of recognizing millions of identities in the wild, this paper focuses on building recognition systems for a small number of people with fewer training images; the recognition system can, for example, build access control systems for research laboratory members or family members. This paper proposes an artificial face image construction method and a half‐face training strategy for transfer learning of pretrained conventional neural network models. Extensive experimental results show that the proposed method improves the performance of state‐of‐the‐art models by utilizing the symmetrical characteristics of human faces. © 2020 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Automated fault detection and diagnosis (AFDD) plays a crucial role in enhancing the energy efficiency of air conditioning systems. Deep learning has emerged as a promising tool for image classification, and its application in the context of AFDD of HVAC systems is gaining traction due to its exceptional performance. However, the deployment cost of deep models in practical scenarios is increased due to the large number of parameters and the lack of interpretability. This paper focuses on improving the potential of deep learning models for AFDD in real HVAC systems. We use pruning to reduce the number of parameters in the model and use layer-wise relevance propagation (LRP) to improve the interpretability of the model. The case study builds a simulation model and 31 kinds of fault data sets based on the actual HVAC in Japan. Based on the findings, Without loss of accuracy, the pruning method can reduce the model size by more than 99 % and maintain 90% classification accuracy. The LRP score allows model users to find out the input data that most affects the results at each diagnosis, improving interpretability. • The study presents a solution for cutting costs of deep learning deployment in HVAC system AFDD. • The method combines pruning and LRP, reducing model size by over 99 • LRP score improves model interpretability by showing impactful input data on diagnosis results. • The study shows the potential of deep learning in HVAC AFDD, enhancing their practicality and usefulness. [ABSTRACT FROM AUTHOR]

Convolutional neural network (CNN), as one of the mainstream deep learning models, has achieved great success in image recognition. All neurons used in CNN are based on the McCulloch‐Pitts model, which is over‐simplified. To further improve CNN's learning capacity, this paper proposes a novel dendritic CNN (DCNN), which considers the nonlinear information processing functions of dendrites in a single neuron. The superiority of DCNN is confirmed based on four widely used image recognition tasks. © 2021 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

A headache and drowsiness are the most common symptoms of fatigue caused by a long duration of work using a visual display terminal (VDT). A sign of the headache generally involves placing a hand on the head, eyes, nose, or face. The recognition of these gestures is a challenging problem due to the difficulty in similar skin color of hands and face. In this paper, a method for classifying six hand over face poses, which can identify the signs of headache for the VDT workers is presented. In the proposed method, a deep learning based on a convolutional neural network (CNN) for the classification of the hand poses is applied. In addition, a class activation map (CAM) to visualize the prediction of the classification network for localization of the hand over face poses was implemented. From the experimental results, the hand poses as the signs of frontal, and unilateral headaches without the classification overfitting and data biasing errors were successfully classified. Our proposed method has achieved high accuracy recognition ratio of 98.5% for classification of the hand over face poses as the prediction of headaches. © 2020 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

This letter presents several combinations of a convolutional neural network (CNN) and its learning method for rotation‐invariant digit recognition. Rotation data augmentation is widely used for improving rotation invariance. Data augmentation commonly assigns the same label to all augmented images of the same source. However, this learning method causes some collisions between original and rotated digits. Thus, this letter presents three types of rotation‐invariance learning methods and applies them to five popular CNN architectures. Experimental results indicate that multi‐task learning on ResNet‐50 is the best combination. © 2020 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Several drowsiness detection technologies have been developed to combat traffic accidents. Drowsiness evaluation has been attempted using the time‐series changes in nasal skin temperature. However, constructing a detection model based on this has been difficult because recent studies have reported an individual difference in skin temperature behaviors. In this study, the conventional method of extracting features was revised. The model of the level of drowsiness was constructed based on facial skin temperature distribution using a convolutional neural network (CNN). The drowsiness level was developed by the New Energy and Industrial Technology Development Organization as an objective drowsiness evaluation index based on facial expressions. With CNNs, features related to learning can be observed as feature quantity distributions. As a result, a general model created has a lack of generality and it is thought that not only the response to drowsiness, but also face shape, exhibit individual differences. Consequently, different features were found in each subject. Through feature maps in individual models, it is believed that skin temperature changes have both reproducible and individual response characteristics to drowsiness, due to the asymmetric left–right change in feature quantity distribution depending on the observed drowsiness level. This method was compared with the conventional method of extracting other features related to drowsiness. It suggested that the skin temperature of not only the nasal region but also the entire face changes as drowsiness increases. Consequently, each discrimination rate calculated by the CNN was at least 20% higher than that obtained via conventional methods. © 2019 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc. [ABSTRACT FROM AUTHOR]

This study was aimed at detecting flower areas in Convallaria keiskei colonies by using unmanned‐aerial‐vehicle (UAV) images. To estimate the number of Convallaria keiskei colonies, we focused on the effective features obtained from the information of flower color and location in the image obtained from the UAV. We proposed a method for estimating the number of colonies by combining image processing and machine learning based on the characteristics of Convallaria keiskei flowers. To evaluate the new effect of the proposed method, we compared the results obtained from the proposed method with those obtained from the comparison method that uses image processing focusing only on brightness and saturation. The results suggest that the proposed method that combines image processing and machine learning is effective for estimating the number of Convallaria keiskei colonies. © 2020 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Segmentation and analysis of individual pores and grains of mudrocks from scanning electron microscope images is non-trivial because of imaging artifacts, variation in pixel grayscale values across images, and overlaps in grayscale values among different physical features such as silt grains, clay grains and pores, which make identifications difficult. Moreover, because grains and pores often have overlapping grayscale values, direct application of threshold-based segmentation techniques is not sufficient. Recent advances in the field of computer vision have made it easier and faster to segment images and identify multiple occurrences of such features in an image, provided that ground-truth data for training the algorithm are available. Here we propose a deep learning SEM image segmentation model, MudrockNet based on Google's DeepLab-v3+ architecture implemented with the TensorFlow library. The ground-truth data were obtained from an image-processing workflow applied to scanning electron microscope images of uncemented muds from the Kumano Basin offshore Japan at depths <1.1 km. The trained deep learning model obtained a pixel-accuracy > 90%, and predictions for the test data obtained a mean intersection over union (IoU) of 0.6663 for silt grains, 0.7797 for clay grains and 0.6751 for pores. We also compared our model with the random forest classifier using trainable Weka segmentation in ImageJ, and it was observed that MudrockNet gave better predictions for silt grains, clay grains and pores in most cases. The size, concentration, and spatial arrangement of the silt and clay grains can affect the petrophysical properties of a mudrock, and an automated method to accurately identify the different grains and pores in mudrocks can help improve reservoir and seal characterization for petroleum exploration and anthropogenic waste sequestration. • MudrockNet allows automated identification of pores and silt in mudrock SEM images. • Model obtained 0.6663(silt), 0.7797(clay) and 0.6751(pores) mean intersection over union on test set. • MudrockNet achieved superior segmentation results than trainable Weka in ImageJ. [ABSTRACT FROM AUTHOR]

• The automatic monitoring of road conditions for multiple countries is addressed. • Deep Learning models are trained for detecting road damages in India, Japan, and Czech. • Recommendations are provided for reusing the data and models released by any country. • A large-scale road damage dataset comprising 26,620 annotated road images is proposed. • The Global Road Damage Detection Challenge'2020 utilizes a part of the proposed data. Many municipalities and road authorities seek to implement automated evaluation of road damage. However, they often lack technology, know-how, and funds to afford state-of-the-art equipment for data collection and analysis of road damages. Although some countries, like Japan, have developed less expensive and readily available Smartphone-based methods for automatic road condition monitoring, other countries still struggle to find efficient solutions. This work makes the following contributions in this context. Firstly, it assesses usability of Japanese model for other countries. Secondly, it proposes a large-scale heterogeneous road damage dataset comprising 26,620 images collected from multiple countries (India, Japan, and the Czech Republic) using smartphones. Thirdly, it proposes models capable of detecting and classifying road damages in more than one country. Lastly, the study provides recommendations for readers, local agencies, and municipalities of other countries when one other country publishes its data and model for automatic road damage detection and classification. A part of the proposed dataset was utilized for Global Road Damage Detection Challenge'2020 and can be accessed at (https://github.com/sekilab/RoadDamageDetector/). [ABSTRACT FROM AUTHOR]

After a major seismic event, structural safety inspections by qualified experts are required prior to reoccupying a building and resuming operation. Such manual inspections are generally performed by teams of two or more experts and are time consuming, labor intensive, subjective in nature, and potentially put the lives of the inspectors in danger. The authors reported previously on the system for a rapid post-earthquake safety assessment of buildings using sparse acceleration data. The proposed framework was demonstrated using simulation of a five-story steel building modeled with three-dimensional nonlinear analysis subjected to historical earthquakes. The results confirmed the potential of the proposed approach for rapid safety evaluation of buildings after seismic events. However, experimental validation on large-scale structures is required prior to field implementation. Moreover, an extension to the assessment of high-rise buildings, such as those commonly used for residences and offices in modern cities, is needed. To this end, a 1/3-scale 18-story experimental steel building tested on the shaking table at E-Defense in Japan is considered. The importance of online model updating of the linear building model used to calculate the Damage Sensitive Features (DSFs) during the operation is also discussed. Experimental results confirm the efficacy of the proposed approach for rapid post-earthquake safety evaluation for high-rise buildings. Finally, a cost-benefit analysis with respect to the number of sensors used is presented. [ABSTRACT FROM AUTHOR]

Nowadays, for numerous reasons, smart farming systems focus on the use of image processing technologies and 5G communications. In this paper, we propose a tracking system for individual cows using an ear tag visual analysis. By using ear tags, the farmers can track specific data for individual cows such as body condition score, genetic abnormalities, etc. Specifically, a four-digit identification number is used, so that a farm can accommodate up to 9999 cows. In our proposed system, we develop an individual cow tracker to provide effective management with real-time upgrading enforcement. For this purpose, head detection is first carried out to determine the cow's position in its related camera view. The head detection process incorporates an object detector called You Only Look Once (YOLO) and is then followed by ear tag detection. The steps involved in ear tag recognition are (1) finding the four-digit area, (2) digit segmentation using an image processing technique, and (3) ear tag recognition using a convolutional neural network (CNN) classifier. Finally, a location searching system for an individual cow is established by entering the ID numbers through the application's user interface. The proposed searching system was confirmed by performing real-time experiments at a feeding station on a farm at Hokkaido prefecture, Japan. In combination with our decision-making process, the proposed system achieved an accuracy of 100% for head detection, and 92.5% for ear tag digit recognition. The results of using our system are very promising in terms of effectiveness. [ABSTRACT FROM AUTHOR]

P-wave first-motion polarity is the most useful information in determining the focal mechanisms of earthquakes, particularly for smaller earthquakes. Algorithms have been developed to automatically determine P-wave first-motion polarity, but the performance level of the conventional algorithms remains lower than that of human experts. In this study, we develop a model of the convolutional neural networks (CNNs) to determine the P-wave first-motion polarity of observed seismic waveforms under the condition that P-wave arrival times determined by human experts are known in advance. In training and testing the CNN model, we use about 130 thousand 250 Hz and about 40 thousand 100 Hz waveform data observed in the San-in and the northern Kinki regions, western Japan, where three to four times larger number of waveform data were obtained in the former region than in the latter. First, we train the CNN models using 250 Hz and 100 Hz waveform data, respectively, from both regions. The accuracies of the CNN models are 97.9% for the 250 Hz data and 95.4% for the 100 Hz data. Next, to examine the regional dependence, we divide the waveform data sets according to the observation region, and then we train new CNN models with the data from one region and test them using the data from the other region. We find that the accuracy is generally high (≳ 95%) and the regional dependence is within about 2%. This suggests that there is almost no need to retrain the CNN model by regions. We also find that the accuracy is significantly lower when the number of training data is less than 10 thousand, and that the performance of the CNN models is a few percentage points higher when using 250 Hz data compared to 100 Hz data. Distribution maps, on which polarities determined by human experts and the CNN models are plotted, suggest that the performance of the CNN models is better than that of human experts. [ABSTRACT FROM AUTHOR]