Reference labels for crop-type mapping are frequently coarse, and administrative land-use registries such as Japan’s eMAF (electronic Map of Agriculture and Forestry) database routinely group agronomically distinct crops under broad, ambiguous categories. This study addresses that problem for the Nakasatsunai region of Hokkaido, Japan, by combining Sentinel-1 synthetic aperture radar (SAR) time series with field-level optical vegetation-index analysis in a modular processing pipeline. The principal novelty of the work is not the pipeline architecture alone but a three-step, Normalized Difference Vegetation Index (NDVI)-driven label-refinement procedure—automatic removal of non-growing or low-amplitude field samples, Euclidean k-means subclass discovery within each coarse label, and trajectory-based label correction—that converts noisy nine-class eMAF labels into a more reliable training set prior to classifier training. The feature set combines the Radar Vegetation Index (RVI), VV and VH backscatter, the γVH/γVV polarization ratio, and NDVI, together with temporal-shape descriptors (phenological timing, peak magnitude, amplitude, maximum slope, and area under the curve) derived from monthly growth trajectories over the 2018 growing season. A Random Forest classifier, together with a gradient-boosting comparator, is evaluated before and after preprocessing under stratified k-fold cross-validation. Across n = 1208 field samples spanning the nine eMAF classes, classification accuracy improved from an overall accuracy of 71.8% on the raw labels to 82.6% after the three-step refinement; Cohen’s kappa increased from 0.63 to 0.77. Correlation analysis indicates that γVH/γVV tracks field-level NDVI more consistently (mean Pearson r = 0.68) than RVI does (mean Pearson r = 0.43) across the eight classes with sufficient samples, motivating its use as a SAR-only phenological proxy; this comparison is extended to the polarimetric PRVI, DPSVI, and DpRVI indices in the discussion. The underlying 80–90% label-accuracy estimate is derived from NDVI trajectory inspection rather than independent, field-surveyed ground truth, and a factorial ablation is used to characterize, to the extent the cross-validated evidence allows, how much of the reported accuracy gain is attributable to label-error correction as opposed to NDVI–SAR feature fusion; both this attribution and the label-accuracy estimate itself are identified as priorities for field validation in future work. The proposed framework is intended to convert coarse, noisy crop labels into a structured and reliable dataset while producing interpretable, field-level phenological insight for agricultural monitoring.

Traditional landslide hazard mapping in Japan relies on labor-intensive field surveys, which are slow, costly, and fail to update dynamically amid rising climate-driven disasters like the 2018 Heavy Rain Event, leaving gaps in timely evacuations. This study addresses these challenges by proposing a semantic segmentation framework using ResUNet to fuse Sentinel-2 optical, Sentinel-1 SAR amplitude, DEM-derived Terrain Ruggedness Index (TRI), and JAXA land cover data, tackling class imbalance with BCE + Dice loss and providing probability/uncertainty maps via 4-TTA for robust hazard delineation under adverse weather. The principal aim is to enable operational, weather-robust hazard zone extraction with AUC upto 0.89 (best multimodal configuration), outperforming single-modality baselines (e.g., optical-only AUC 0.74; SAR-only 0.69) through synergistic feature fusion, while highlighting multimodal SAR's edge for cloud-obscured scenarios. Validated on Hiroshima Prefecture data-Japan's highest-risk region with similar to 32,000 hazard spots-this approach demonstrates pre/post-disaster change detection, but reveals limitations in spatial generalization due to region-specific training.

This study proposes an earthquake disaster detection method based on interferometric synthetic aperture radar (InSAR) using synthetic pre-disaster SAR data generated from optical satellite images. Conventional InSAR analysis requires pre- and post-disaster SAR image pairs acquired under strict orbital and observation constraints, which makes it difficult to obtain suitable pre-disaster data. In the proposed approach, a digital elevation model (DEM) and land-cover information are combined with optical imagery, and generative adversarial networks (GANs), specifically pix2pixHD and CycleGAN, are used to generate pseudo-SAR data that include both amplitude and phase components. Experimental results using Sentinel-1 SAR and Sentinel-2 multispectral instrument (MSI) data demonstrate that pix2pixHD achieves higher conversion accuracy than CycleGAN, with a peak signal-to-noise ratio (PSNR) of 21.25 dB and a histogram intersection of 65.25%, and that the generated pre-disaster SAR images can be interfered with post-disaster SAR observations to detect earthquake-induced surface changes in the 2024 Noto Peninsula event. These findings indicate that the proposed method can extend the applicability of InSAR to areas and events where suitable pre-disaster SAR acquisitions are unavailable, contributing to rapid earthquake disaster assessment.

Ship monitoring using Synthetic Aperture Radar (SAR) data faces significant challenges in detecting small vessels due to low spatial resolution and speckle noise. While ESRGAN (Enhanced Super-Resolution Generative Adversarial Network) has shown promise for image super-resolution, it struggles with SAR imagery characteristics. This study proposes SA/SU-ESRGAN, which extends the SU-ESRGAN framework by incorporating a spatial attention mechanism loss function. SU-ESRGAN introduced semantic structural loss to accurately preserve ship shapes and contours; our enhancement adds spatial attention to focus reconstruction efforts on ship regions while suppressing background noise. Experimental results demonstrate that SA/SU-ESRGAN successfully detects small vessels that remain undetectable by SU-ESRGAN, achieving improved detection capabilities with a PSNR of approximately 26 dB (SSIM is around 0.5) and enhanced visual clarity in ship boundaries. The spatial attention mechanism effectively reduces noise influence, producing clearer super-resolution results suitable for maritime surveillance applications. Based on the HRSID dataset, a representative dataset for evaluating ship detection performance using SAR data, we evaluated ship detection performance using images in which the spatial resolution of the SAR data was artificially degraded using a smoothing filter. We found that with a 4 × 4 filter, all eight ships were detected without any problems, but with an 8 × 8 filter, only three of the eight ships were detected. When super-resolution was applied to this, six ships were detected.

Continuous vegetation monitoring is essential for predicting crop varieties and yields; however, optical satellite data are frequently unavailable due to cloud cover. To overcome this limitation, this study proposes a method for generating pseudo-NDVI (Normalized Difference Vegetation Index) imagery from RVI (Radar Vegetation Index) derived from Synthetic Aperture Radar (SAR) data using Generative Adversarial Networks (GANs). Two architectures—pix2pixHD (supervised) and CycleGAN (unsupervised)—were evaluated using Sentinel-1 and Sentinel-2 data under identical conditions. By introducing RVI as an intermediate feature instead of directly converting SAR backscatter to NDVI, the proposed method enhanced physical interpretability and improved correlation with NDVI. Quantitative results show that pix2pix achieved higher accuracy (SSIM = 0.5667, PSNR = 22.24 dB, RMSE = 20.54) than CycleGAN (SSIM = 0.5240, PSNR = 19.54 dB, RMSE = 28.02), with further improvement when combining VV and VH polarization data. Although the absolute accuracy remains moderate, this approach enables continuous annual NDVI time series reconstruction for crop monitoring under persistent cloud conditions, demonstrating clear advantages over conventional direct SAR-to-NDVI conversion methods.

Glaucoma is a leading cause of blindness, and early detection is critical. Convolutional neural networks (CNNs) have shown impressive performance in glaucoma diagnosis, but their black-box nature remains a barrier to clinical use. Existing explainable AI (XAI) methods such as Grad-CAM have limitations in identifying and quantifying subtle regional features. In this study, we propose a method to clarify what CNNs focus on by analyzing how model performance changes under localized adversarial noise. Using VGG16 for glaucoma classification, we applied noise generated by the Fast Gradient Sign Method (FGSM) to the whole fundus image and to specific subregions, then compared the impact on classification performance. Results showed that perturbations to the optic disc, especially its outer margin, had the greatest effect on model performance. This suggests that the CNN captures fine anatomical features such as optic disc cupping and neuroretinal rim thinning, which aligns with what ophthalmologists typically look for. At the same time, perturbations in the macula and perivascular regions also affected performance, indicating gaps between current clinical diagnostic criteria and the CNN’s decision-making process. This approach can help establish the clinical reliability of CNNs and may also reveal features that have not been recognized in conventional clinical practice.

This study investigates the use of the radar vegetation index (RVI) derived from Sentinel-1 synthetic aperture radar (SAR) data for landslide detection. Traditional landslide detection methods often rely on the Normalized Difference Vegetation Index (NDVI) derived from optical imagery, which is susceptible to limitations imposed by weather conditions (clouds, rain) and nighttime. In contrast, SAR data, acquired by Sentinel-1, provides all-weather, day-and-night coverage. To leverage this advantage, we propose a novel approach utilizing RVI, a vegetation index calculated from SAR data, to identify non-vegetated areas, which often indicate potential landslide zones. To enhance the accuracy of non-vegetated area classification, we employ the high-performing EfficientNetV2 deep learning model. We evaluated the classification performance of EfficientNetV2 using RVI derived from Sentinel-1 SAR data with VV and VH polarizations. Experiments were conducted on SAR imagery of the Iburi district in Hokkaido, Japan, severely impacted by an earthquake in 2018. Our findings demonstrate that the classification performance using RVI with both VV and VH polarizations significantly surpasses that of using VV and VH polarizations alone. These results highlight the effectiveness of RVI for identifying non-vegetated areas, particularly in landslide detection scenarios. The proposed RVI-based method has broader applications beyond landslide detection, including other disaster area assessments, agricultural field monitoring, and forest inventory.

The strawberry industry faces significant challenges such as harvest inefficiency and inconsistent quality due to labor shortages and reliance on inexperienced workers. One factor contributing to this is the lack of quantified criteria for strawberry harvesting decisions. In this study, we propose a harvest support system that uses smart glasses as a medium to provide real-time feedback on quantified ripeness and size information through computer vision technology. In the experiment, we compared the efficiency and accuracy of harvesting between cases where humans made harvesting decisions alone and cases where they were supported by the developed system to verify the usefulness of the developed system. The results showed that the system improved the harvesting speed by 17%, the accuracy of size evaluation by 25% and ripeness evaluation by 8% compared to the case where only human judgment was used. These results suggest that this system not only improves productivity and ensures consistent quality but also is expected to increase the number of new entrants in the strawberry industry, potentially addressing the labor shortage in the strawberry sector.

One factor that has a positive correlation with the risk of traffic accidents is the pain experienced by drivers. This pain is sometimes expressed facially by the driver and can be subjectively perceived by others. By observing the facial expression of drivers, it can estimate the pain experienced at that point in time and intervene to prevent some accidents. A method to automatically estimate the pain level expressed by a driver using their facial expression will be proposed in this study. The model is trained by a convolution neural network based on a public dataset of facial expressions at various pain levels. This model is then used to automatically classify the pain level perceived using only the facial expressions of drivers. The result of the automated classification is then compared to ratings of subjective feelings of the driver's pain evaluated by a medical doctor. The experiment results showed that the model classified the pain level expressed facially by the drivers matched that of the classification by the medical doctor at a rate of 80%.

The integration of modern manufacturing systems has promised increased flexibility, productivity, and efficiency. In such an environment, collaboration between humans and robots in a shared workspace is essential to effectively accomplish shared tasks. Strong communication among partners is essential for collaborative efficiency. This research investigates an approach to non-verbal communication cues. The system focuses on integrating human motion detection with vision sensors. This method addresses the bias human action detection in frames and enhances the accuracy of perception as information about human activities to the robot. By interpreting spatial and temporal data, the system detects human movements through sequences of human activity frames while working together. The training and validation results confirm that the approach achieves an accuracy of 91

Individuals with visual impairments often rely on assistive tools such as white canes and guide dogs to navigate their environments. While these tools provide a certain level of support, their effectiveness is frequently constrained in complex or dynamically changing environments, even with extensive user training. To address these limitations, we have developed a smart navigation robot that integrates artificial intelligence for object detection, offering a viable alternative to traditional assistive tools. The robot is designed to provide real-time assistance through auditory alerts, all while allowing the user to maintain full control over the robot’s direction according to their intentions. The robot’s effectiveness was evaluated through an experimental study in which participants navigated diverse environments using both the smart navigation robot and a white cane. Participant perceptions of the robot’s usability, reliability, safety, and interaction quality were evaluated using the Godspeed Questionnaire Series. The comparative analysis revealed that the smart navigation robot outperformed the white cane, particularly in dynamic scenarios. These findings suggest that the robot has the potential to substantially improve the quality of life and independence of individuals with visual impairments, offering a greater degree of freedom than was previously attainable.

In recent years, advancements in image classification technology have led to significant improvements in classification accuracy. However, challenges remain, such as difficulties in image classification when redundant information is present, even when using state-of-the-art deep learning methods, and the need for large amounts of training samples for deep learning models. To address these issues, we proposes a method that enhances color information for image classification by combining multi-illumination and multispectral cameras, and utilizes log-linearized gaussian mixture neural network that can classify images with a small number of training samples. The proposed system utilizes a multi-spectral camera capable of capturing Red (R), Green (G), Blue (B), and Near-Infrared (NIR) images (RGB-NIR), along with corresponding multi-spectral illumination. By enhancing color information and clarifying differences in object features, this approach enables high-accuracy color classification with an unprecedentedly small dataset when input into the log-linearized gaussian mixture neural network. Experimental results demonstrated the effectiveness of the proposed system, achieving 100% color classification accuracy on green tea samples with similar color features. This achievement is expected to contribute to various fields such as manufacturing, healthcare, chemistry, and agriculture, where multispectral imaging is increasingly utilized.

Currently, 35.2% of mothers suffer from hand and wrist pain after giving birth. Their physical problems are often related to the posture in which they carry their baby. Therefore, correct posture when carrying the baby is important to avoid postpartum physical problems such as tendonitis. However, determining the infant-carrying posture requires feedback from an expert, (e.g., a midwife), and is time-consuming. To overcome this problem, an infant-carrying posture determination (ICPD) method is proposed using an RGB camera and the BlazePose pose estimation model. With the ICPD method, a person carrying an infant can easily determine the quality of their posture when carrying an infant. To achieve a more accurate determination, the ICPD method normalizes the infant-carrying posture and selects features based on the area under the receiver operating characteristic curve, which is a widely used performance measure in classification models. The postures of 28 mothers while carrying infants was experimentally determined to validate the proposed system. The experimental results confirmed that ICPD was more accurate on the test dataset than conventional methods, both with and without feature selection.

As technology develops and robots gradually become a part of everyday life, devices that provide physical assistance to humans will become increasingly common. Such robots must often physically interact with the user for extended periods of time and must provide auxiliary forces that complement the user's strength to accomplish certain goals. This requires close cooperation and close coordination between the human user and the assistive robot. The purpose of this study is to understand how two humans cooperate and coordinate in order to gain insight into how to improve human-robot coordination. In this study, a tracking task involving a pair of humans was conducted. The task used “predictable” and “unpredictable” sine waves. In the experiment, the roles of the two humans were divided into a “user” focusing on the experimental task and an “assistant” focusing on support, and the amount of information given to the “assistant” was divided into three levels (“Full knowledge”, “Short-term knowledge”, and “No knowledge”) to try to understand the coordination and cooperation tendencies of the two humans. The experimental results showed that in some conditions, the rate of increase in error and load decreased as the amount of task knowledge given to the “assistant” increased, compared to when the task was performed “solo”.

The estimation of leaf area is an important measure for understanding the growth, development, and productivity of tomato plants. In this study, we focused on the leaf area of a potted tomato plant and proposed methods, namely, NP, D2, and D3, for estimating its leaf area. In the NP method, we used multiple tomato plant images from different viewing angles to reduce the estimation error of the leaf area, whereas in the D2 and D3 methods, we further compensated for the perspective effects. The performances of the proposed methods were experimentally assessed using 40 “Momotaro Peace” tomato plants. The experimental results confirmed that the NP method had a smaller mean absolute percentage error (MAPE) on the test set than the conventional estimation method that uses a single tomato plant image. Likewise, the D2 and D3 methods had a smaller MAPE on the test set than the conventional method that did not compensate for perspective effects.

Method for earthquake disaster area detection with just a single satellite-based SAR data which is acquired on the day after earthquake based on object detection method of YOLOv8 and Detectron2 is proposed. Through experiments with several SAR data derived from the different SAR satellites which observed Noto Peninsula earthquake occurred on the first of January 2024, it is found that the proposed method works well to detect several types of damages effectively. Also, it is found that the proposed method based on "Roboflow" and YOLOv8 as well as Detectron2 for annotation and object detection is appropriate for disaster area detection. Furthermore, it is possible to detect disaster areas even if just one single SAR data which acquired on the day after the disaster occurred because the trained learning model for disaster area detection is created through experiments.

As people age, their ability to balance declines. This leads to an increased risk of falls, which often results in loss of independence due to the need for nursing care. The earlier we are able to detect this decline and appropriately assess the risk, the better we will be able to mitigate the risk of falling. To date, however, visual control of balance is often only assessed under two conditions: with the eyes open and closed. Understanding how human postural balance changes when more complex visual stimuli are applied can provide a more detailed picture of a person's balance capability. This study aims to develop a fall risk assessment system for the elderly using visual stimuli as a new indicator for balance assessment. To this end, the effects of various parameters of visual stimuli on postural control using a VR headset were investigated. The results showed that when oscillating a virtual scene, the mediolateral oscillations tended to have less influence than the oscillations in the other direction, while the increasing frequency of the oscillations resulted in more postural sway in the anteroposterior direction. However, increasing amplitude appears to increase postural sway in both the anteroposterior and mediolateral directions. Our results suggest that scene oscillation in a VR headset may be able to provide a more detailed understanding of a person's balance capability; however, more work is required to verify these effects.

A method for landslide area detection based on EfficientNetV2 with optical image converted from SAR image using pix2pixHD with a spatial attention mechanism in the loss function is proposed. Meteorological landslides such as landslides after heavy rains occur regardless of day or night and weather conditions. Meteorological landslides such as landslides are easier to visually judge using optical images than SAR images, but optical images cannot be observed at night, in the rain, or on cloudy days. Therefore, we devised a method to convert SAR images, which allow all-weather observation regardless of day or night, into optical images using pix2pixHD, and to learn about landslide areas using the converted optical images to build a trained model. We used SAR and optical images derived from Sentinel-1 and -2, which captured landslides caused by the earthquake on 14 April 2016, as training data, and constructed a learning model that classifies landslide areas using EfficientNetV2. We evaluated the superiority of the proposed method by comparing it with a learning model that uses only SAR images. As a result, it was confirmed that the F1-score and AUC were 0.3396 and 0.2697, respectively, when using only SAR images, but were improved by 1.52 to 1.84 times to 0.6250 and 0.4109, respectively, when using the proposed method.

The estimation of leaf area is an important measurement in understanding the growth, development, and productivity of tomato plants. In this study, we focus on the leaf area of each tomato plant in a greenhouse environment and propose a method for estimating its leaf area using multiple RGB-D cameras, which we call D 4 method. In the D 4 method, we use instance segmentation and multiple RGB-D cameras to estimate the leaf area of each tomato plant in the greenhouse environment. The performance of the D 4 method is assessed in an experiment with 30 "Momotaro Peace" tomato plants. This experiment highlights that the D 4 method has a mean absolute percentage error on a test set of 10.1%, which has a better performance than the conventional other methods.

In this study, we experimentally analyzed whether people can perceive the accuracy or smartness of AI judgments, and whether the judgment accuracy of AI and the level of confidence in those judgments affect people’s decision-making. The results showed that people may perceive an AI’s smartness even when it only presents information on the results of its judgments. The results also suggest that AI accuracy and confidence affect human decision-making, and that the magnitude of the effect of AI confidence varies with AI accuracy. We also found that when a person’s ability to make a decision is less than or equal to the AI’s ability to make a decision, the human performance in a binary decision task improves regardless of AI accuracy. The results obtained in this study are similar in some respects to relationships in which people make decisions while interacting, and the findings from research on human interactions may apply to the research and development of human–AI interactions.