BiFormer Attention‐Guided Multiscale Fusion Mask2former Networks for Fish Abnormal Behavior Recognition and Segmentation DOI Creative Commons

J. Z. Liu,

Zeyuan Hu,

Yixi Zhang

и другие.

Aquaculture Research, Год журнала: 2024, Номер 2024(1)

Опубликована: Янв. 1, 2024

To address the issues of accurately identifying and tracking individual fish abnormal behaviors poor adaptability in aquaculture field, this paper proposes a Mask2former model combined with bidirectional routing attention mechanism (BiFormer) multiscale dilated (MSDA) module for behavior recognition segmentation. compensate lack publicly available datasets on behavior, we created “FISH_segmentation_2023” dataset, which includes four types behaviors. First, by introducing BiFormer mechanism, can better capture critical temporal spatial information image sequences, significantly enhancing feature representation. Second, after processing maps pixel decoder, MSDA is introduced to perform fusion these features. The fused features are then passed transformer further model’s ability recognize Finally, improve performance class imbalance designed composite loss function combining focal dice (FD loss). This balance influence easy difficult‐to‐classify samples while optimizing segmentation performance, thereby improving accuracy mean intersection over union (mIoU) metrics. Experimental results show that FD (BMF)‐Mask2former exhibits high achieving average (IoU), accuracy, recall values 92.33%, 95.63%, 94.82%, respectively, self‐built FISH_segmentation_2023 representing improvements 6.10%, 4.50%, 5.09%, compared model. study demonstrates proposed both local contextual through methods, resulting high‐quality outcomes.

Язык: Английский

YOLOv10-UDFishNet: detection of diseased Takifugu rubripes juveniles in turbid underwater environments DOI

Wan Tu,

Hong Liang Yu, Zijian Wu

и другие.

Aquaculture International, Год журнала: 2025, Номер 33(1)

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

1
FeYOLO: Improved YOLOv7-tiny model using feature enhancement modules for the detection of individual silkworms in high-density and compact conditions DOI

Hongkang Shi,

Linbo Li,

Shiping Zhu

и другие.

Computers and Electronics in Agriculture, Год журнала: 2025, Номер 231, С. 109966 - 109966

Опубликована: Янв. 22, 2025

Язык: Английский

Процитировано

0
Picking point localization method of table grape picking robot based on you only look once version 8 nano DOI
Yanjun Zhu,

Shunshun Sui,

Wensheng Du

и другие.

Engineering Applications of Artificial Intelligence, Год журнала: 2025, Номер 146, С. 110266 - 110266

Опубликована: Фев. 15, 2025

Язык: Английский

Процитировано

0
Real-time detection of hypoxic stress behavior in aquaculture fish using an enhanced YOLOv8 model DOI
Changjiang Cai,

Shaohui Tan,

Xinmiao Wang

и другие.

Aquaculture International, Год журнала: 2025, Номер 33(3)

Опубликована: Фев. 25, 2025

Язык: Английский

Процитировано

0
Accurate machine vision identification of GCHD symptom using a self-attention-based CNN model with adaptive fish separation DOI Creative Commons
Xiang Shen, Zehui Liu,

Wei Qin

и другие.

Smart Agricultural Technology, Год журнала: 2025, Номер unknown, С. 100871 - 100871

Опубликована: Фев. 1, 2025

Язык: Английский

Процитировано

0
FQDNet: A Fusion-Enhanced Quad-Head Network for RGB-Infrared Object Detection DOI Creative Commons
Fangzhou Meng,

Aoping Hong,

Hongying Tang

и другие.

Remote Sensing, Год журнала: 2025, Номер 17(6), С. 1095 - 1095

Опубликована: Март 20, 2025

RGB-IR object detection provides a promising solution for complex scenarios, such as remote sensing and low-light environments, by leveraging the complementary strengths of visible infrared modalities. Despite significant advancements, two key challenges remain: (1) effectively integrating multi-modal features within lightweight frameworks to enable real-time performance (2) fully utilizing multi-scale features, which are crucial detecting objects varying sizes but often underexploited, leading suboptimal accuracy. To address these challenges, we propose FQDNet, novel network that integrates an optimized fusion strategy with Quad-Head framework. enhance feature fusion, introduce Channel Swap SCDown Block (CSSB) initial interaction Spatial Attention Fusion Module (SCAFM) further refine integration features. improve utilization, designed Dynamic-Weight-based Detector (DWQH), dynamically assigns weights different scales, enabling adaptive enhancing representation. This mechanism significantly improves performance, particularly small objects. Furthermore, ensure applicability, incorporate optimizations, including Partial Cross-Stage Pyramid (PCSP) modules, reduce computational complexity while maintaining high FQDNet was evaluated on three public datasets—M3FD, VEDAI, LLVIP—achieving mAP@[0.5:0.95] gains 4.4%, 3.5%, 3.1% over baseline, only 0.4 M increase in parameters 5.5 GFLOPs overhead. Compared state-of-the-art algorithms, our method strikes better balance between accuracy efficiency exhibiting strong robustness across diverse scenarios.

Язык: Английский

Процитировано

0
TSSSKD-YOLO: an intelligent classification and defect detection method of insulators on transmission lines by fusing knowledge distillation in multiple scenarios DOI
Yongsheng Ye, Gary Tan, Qiang Liu

и другие.

Multimedia Systems, Год журнала: 2025, Номер 31(3)

Опубликована: Апрель 7, 2025

Язык: Английский

Процитировано

0
DF-DETR: Dead fish-detection transformer in recirculating aquaculture system DOI
Tao Fu, Dejun Feng, Pingchuan Ma

и другие.

Aquaculture International, Год журнала: 2024, Номер 33(1)

Опубликована: Ноя. 13, 2024

Язык: Английский

Процитировано

1
BiFormer Attention‐Guided Multiscale Fusion Mask2former Networks for Fish Abnormal Behavior Recognition and Segmentation DOI Creative Commons

J. Z. Liu,

Zeyuan Hu,

Yixi Zhang

и другие.

Aquaculture Research, Год журнала: 2024, Номер 2024(1)

Опубликована: Янв. 1, 2024

To address the issues of accurately identifying and tracking individual fish abnormal behaviors poor adaptability in aquaculture field, this paper proposes a Mask2former model combined with bidirectional routing attention mechanism (BiFormer) multiscale dilated (MSDA) module for behavior recognition segmentation. compensate lack publicly available datasets on behavior, we created “FISH_segmentation_2023” dataset, which includes four types behaviors. First, by introducing BiFormer mechanism, can better capture critical temporal spatial information image sequences, significantly enhancing feature representation. Second, after processing maps pixel decoder, MSDA is introduced to perform fusion these features. The fused features are then passed transformer further model’s ability recognize Finally, improve performance class imbalance designed composite loss function combining focal dice (FD loss). This balance influence easy difficult‐to‐classify samples while optimizing segmentation performance, thereby improving accuracy mean intersection over union (mIoU) metrics. Experimental results show that FD (BMF)‐Mask2former exhibits high achieving average (IoU), accuracy, recall values 92.33%, 95.63%, 94.82%, respectively, self‐built FISH_segmentation_2023 representing improvements 6.10%, 4.50%, 5.09%, compared model. study demonstrates proposed both local contextual through methods, resulting high‐quality outcomes.

Язык: Английский

Процитировано

0