ZZZhen Zhang

Zhen Zhang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Zhen Zhang has a total of 13 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Mitigating Cross-Dataset Feature Shifts via a Multi-Loss Self-Reconstruction Filter: A Correction-Detection Framework for UAV-Based Yuzu Detection

    by Zhen Zhang, Yuu Tanimoto, Makoto Iwata, Shinichi Yoshida

    Deep learning-based yuzu detection systems often suffer performance degradation when the training and inference datasets differ in collection sites, acquisition years, and imaging conditions. To address this issue, this study proposes a multi-loss self-reconstruction (MLSR) filter and an MLSR-based correction-detection framework.The proposed MLSR filter is an encoder-decoder architecture, where the input image is encoded into feature representations and then decoded back into the image space. It aims to reconstruct natural image characteristics while suppressing detection-irrelevant variations. To achieve this objective, we propose a staged optimization strategy with multi-loss constraints to train the filter, in which L1 and SSIM losses constrain the reconstruction to recover the basic image structure, the VGG perceptual loss constrains it to preserve detection-relevant features, and the PatchGAN adversarial loss prevents the reconstructed images from deviating excessively from natural image characteristics. In the correction-detection framework, the trained MLSR filter is applied to both training and inference datasets before yuzu detection. The proposed correction-detection framework was evaluated on UAV (Unmanned Aerial Vehicle) imagery collected under different orchards, years, and illumination conditions. The results showed consistent improvements for YOLOv5, YOLOv8, and YOLOv11, with maximum gains of 5.62% in F1-score, 7.72% in mAP50, and 6.95% in mAP50-90. We argued that MLSR filtering mitigated feature shifts between the training and inference datasets, allowing YOLO to capture more consistent patterns across them and thereby improving detection performance. However, this mitigation was reflected in YOLO backbones but did not introduce obvious visual changes. Therefore, we measured the similarity between the training and inference datasets based on YOLO backbone representations before and after MLSR filtering. The results showed that YOLO perceived higher feature similarity between the datasets at detection-relevant scales after MLSR filtering, consistent with the reported detection improvements.

  • Open Access

    Design Tradeoffs in Asynchronous BD Pipeline Styles of RISC-V: An FPGA-Based Comparative Study

    by Valeeprakhon Tamnuwat, Zhen Zhang, Makoto Iwata

    Asynchronous processors adopt diverse pipeline styles, including linear, frequency-adaptive, selective, reduced, and combined styles, to balance performance, energy, and area. However, prior studies remain fragmented because benchmarks, tool support, and design flows differ. Those make fair comparison difficult and prevent a clear understanding of the trade-offs required for choosing pipeline styles for specific applications. Moreover, existing delay-placement flows rely on iterative synthesis or manual engineering change orders (ECOs), which are time-consuming, error-prone, and inefficient. To overcome these limitations, we introduce a morphological delay-placement constraint (MDPC), an automated delay-placement flow for bundled-data (BD) circuits on FPGAs. Compared with conventional flows, MDPC achieves single-pass timing closure with reduced slack deviation by up to 31.66% and LUT utilization by up to 5.56X, thereby enhancing performance by up to 11.94% and energy efficiency by up to 10.42%. Using MDPC, we systematically evaluate five asynchronous RISC-V pipeline styles against a synchronous baseline with typical benchmarks. The frequency-adaptive pipeline achieves the highest speed gains of 1.11%. The selective pipeline delivers the strongest energy savings of up to 65.85%. The reduced pipeline optimizes resource utilization, lowering area by up to 3.12%. The combined pipeline provides moderate improvements with energy reductions of up to 64.40%. In contrast, the linear pipeline exhibits the lowest overall efficiency, with execution slowdowns of up to 37.76%. The results reveal distinct trade-offs: the frequency-adaptive pipeline maximizes speed, the selective pipeline minimizes energy, the reduced pipeline achieves superior resource efficiency, the combined pipeline offers moderate overall gains, while the linear pipeline remains the least competitive.

  • Open Access

    Object Detection for Yellow Maturing Citrus Fruits from Constrained or Biased UAV Images: Performance Comparison of Various Versions of YOLO Models

    by Yuu Tanimoto, Zhen Zhang, Shinichi Yoshida

    Citrus yield estimation using deep learning and unmanned aerial vehicles (UAVs) is an effective method that can potentially achieve high accuracy and labor savings. However, many citrus varieties with different fruit shapes and colors require varietal-specific fruit detection models, making it challenging to acquire a substantial number of images for each variety. Understanding the performance of models on constrained or biased image datasets is crucial for determining methods for improving model performance. In this study, we evaluated the accuracy of the You Only Look Once (YOLO) v8m, YOLOv9c, and YOLOv5mu models using constrained or biased image datasets to obtain fundamental knowledge for estimating the yield from UAV images of yellow maturing citrus (Citrus junos) trees. Our results demonstrate that the YOLOv5mu model performed better than the others based on the constrained 25-image datasets, achieving a higher average precision at an intersection over union of 0.50 (AP@50) (85.1%) than the YOLOv8m (80.3%) and YOLOv9c (81.6%) models in the training dataset. On the other hand, it was revealed that the performance improvement due to data augmentation was high for the YOLOv8m and YOLOv9c models. Moreover, the impact of the bias in the training dataset, such as the light condition and the coloring of the fruit, on the performance of the fruit detection model is demonstrated. These findings provide critical insights for selecting models based on the quantity and quality of the image data collected under actual field conditions.


  • Open Access

    Multi-pooling 3D Convolutional Neural Network for fMRI Classification of Visual Brain States

    by Zhen Zhang, Masaki Takeda, Makoto Iwata

    Neural decoding of visual object classification via functional magnetic resonance imaging (fMRI) data is challenging and is vital to understand underlying brain mechanisms. This paper proposed a multi-pooling 3D convolutional neural network (MP3DCNN) to improve fMRI classification accuracy. MP3DCNN is mainly composed of a three-layer 3DCNN, where the first and second layers of 3D convolutions each have a branch of pooling connection. The results showed that this model can improve the classification accuracy for categorical (face vs. object), face sub-categorical (male face vs. female face), and object sub-categorical (natural object vs. artificial object) classifications from 1.684% to 14.918% over the previous study in decoding brain mechanisms.

  • Open Access

    A Novel Multimodule Neural Network for EEG Denoising

    by Zhen Zhang, Xiaoyan Yu, Xianwei Rong, Makoto Iwata

    In this paper, a novel multi-module neural network (MMNN) is proposed to remove ocular artifacts (OAs) and myogenic artifacts (MAs) from noisy single-channel electroencephalogram (EEG) signals. This network is a based on deep learning (DL) architecture consisting of multiple denoising modules connected in parallel. Each denoising module is built using one-dimensional convolutions (Conv1Ds) and fully connected (FC) layers, and it estimates not only clean EEG signals but also artifacts. The proposed MMNN has two main advantages. Frist, the multiple denoising modules can purify noisy input EEG signals by continuously removing artifacts in the forward propagation. Second, the parallel architecture allows the parameters of each denoising module to be updated concurrently in the backpropagation, thereby improving the learning capacity of neural networks. We tested the network denoising performance using a recent public database, namely, EEGdenoiseNet. The results revealed that the proposed network reduced the temporal relative root mean square error (T-RRMSE) and spectral relative root mean square error (S-RRMSE) by at least 6% and enhanced the correlation coefficient (CC) by at least 3% over the state-of-the-art approaches. These significant performance improvements were confirmed by observing the deviation distribution between the denoised and clean signals. Furthermore, the proposed network achieved a similar performance efficiency with only 60% of the training data compared to the existing DL models.


  • Open Access

    Spatial-Temporal Neural Network for P300 Detection

    by Zhen Zhang, Xiaoyan Yu, Xianwei Rong, Makoto Iwata

    P300 spellers are common brain-computer interface (BCI) systems designed to transfer information between human brains and computers. In most P300 detections, the P300 signals are collected by averaging multiple electroencephalographic (EEG) changes to the same target stimuli, so the participants are obliged to endure multiple repeated stimuli. In this study, a spatial-temporal neural network (STNN) based on deep learning (DL) is proposed for P300 detection. It detects P300 signals by combining the outputs from a temporal unit and a spatial unit. The temporal unit is a flexible framework consisting of several temporal modules designed for analyzing brain potential changes in the time domain. The spatial unit combines one-dimensional convolutions (Conv1Ds) and linear layers to generalize P300 features from the space domain, and it can decode EEG signals recorded using different numbers of electrodes. Both amyotrophic lateral sclerosis (ALS) patients and healthy subjects can benefit from this study. In the within-subject P300 detection and the cross-subject P300 detection, our approach gained higher performance with fewer repeated stimuli than other comparative approaches. Furthermore, we applied the proposed STNN in the P300 detection challenge of BCI Competition III. The accuracy score was 89% in the fifth round of repeated stimuli, outperforming the best result in the literature (accuracy = 80%) to the best of our knowledge. The results demonstrate that the proposed STNN performs well with limited stimuli and is robust enough for various P300 detections.