Hao Peng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Hao Peng has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Yi Wang
College of Information and Intelligence, Hunan Agricultural University, Changsha 410128, China
9 shared publications - Cheng Ouyang
College of Information and Intelligence, Hunan Agricultural University, Changsha 410128, China
6 shared publications - Wenwu Hu
College of Mechanical and Electrical Engineering, Hunan Agricultural University, Changsha 410128, China
6 shared publications - Ping Jiang
School of Mechanical and Electrical Engineering, Hunan Agricultural University, Changsha 410128, China
5 shared publications - Hailin Chen
College of Information and Intelligence, Hunan Agricultural University, Changsha 410128, China
4 shared publications - Jialiang Peng
College of Information and Intelligence, Hunan Agricultural University, Changsha 410128, China
4 shared publications
Latest Publications
- Open Access
MECT-MobileViT: A Lightweight Fish Weight Prediction Model Based on Dual-View Morphological Feature Fusion and Anti-Interference Attention
In intensive aquaculture, non-invasive real-time monitoring of morphological traits and body weight of largemouth bass (Micropterus salmoides) is essential for precision feeding and yield estimation. Manual measurement is laborious and stressful, whereas vision-based methods are challenged by insufficient dual-view feature fusion, poor robustness to underwater noise, and over-parameterized models unsuitable for edge deployment. To address these issues, a lightweight framework, MECT-MobileViT, is proposed based on MobileViT-xxs. A Morphometric-Guided Multi-Scale Fusion module is designed to couple physical priors with dual-branch visual features, strengthening shape–weight association. An ECA-NL attention block employing instance normalization, GLU gating, and threshold filtering is embedded to enhance feature robustness against visual disturbances typical in aquaculture and to accentuate critical morphological features. A three-stage synergistic pruning strategy—attention head pruning, structured channel pruning, and depthwise separable attention substitution—is applied to achieve substantial compression while preserving representational capacity. Experiments on a self-built lateral–dorsal dual-view dataset show that the proposed model significantly outperforms mainstream benchmarks. The pruned version attains an R2 of 0.8266 and an RMSE of 16.4201, with less than 2% accuracy degradation relative to the best unpruned model, and contains only 7.34 M parameters. This study demonstrates a promising prototype for contactless, stress-free weight estimation in largemouth bass and offers new technical insights into feature fusion, noise suppression, and collaborative model compression for aquaculture visual perception.
Open AccessMECT-MobileViT: A Lightweight Fish Weight Prediction Model Based on Dual-View Morphological Feature Fusion and Anti-Interference Attention
In intensive aquaculture, non-invasive real-time monitoring of morphological traits and body weight of largemouth bass (Micropterus salmoides) is essential for precision feeding and yield estimation. Manual measurement is laborious and stressful, whereas vision-based methods are challenged by insufficient dual-view feature fusion, poor robustness to underwater noise, and over-parameterized models unsuitable for edge deployment. To address these issues, a lightweight framework, MECT-MobileViT, is proposed based on MobileViT-xxs. A Morphometric-Guided Multi-Scale Fusion module is designed to couple physical priors with dual-branch visual features, strengthening shape–weight association. An ECA-NL attention block employing instance normalization, GLU gating, and threshold filtering is embedded to enhance feature robustness against visual disturbances typical in aquaculture and to accentuate critical morphological features. A three-stage synergistic pruning strategy—attention head pruning, structured channel pruning, and depthwise separable attention substitution—is applied to achieve substantial compression while preserving representational capacity. Experiments on a self-built lateral–dorsal dual-view dataset show that the proposed model significantly outperforms mainstream benchmarks. The pruned version attains an R2 of 0.8266 and an RMSE of 16.4201, with less than 2% accuracy degradation relative to the best unpruned model, and contains only 7.34 M parameters. This study demonstrates a promising prototype for contactless, stress-free weight estimation in largemouth bass and offers new technical insights into feature fusion, noise suppression, and collaborative model compression for aquaculture visual perception.
- Open Access
ECR-MobileNet: An Imbalanced Largemouth Bass Parameter Prediction Model with Adaptive Contrastive Regression and Dependency-Graph Pruning
The precise, non-destructive monitoring of fish length and weight is a core technology for advancing intelligent aquaculture. However, this field faces dual challenges: traditional contact-based measurements induce stress and yield loss. In addition, existing computer vision methods are hindered by prediction biases from imbalanced data and the deployment bottleneck of balancing high accuracy with model lightweighting. This study aims to overcome these challenges by developing an efficient and robust deep learning framework. We propose ECR-MobileNet, a lightweight framework built on MobileNetV3-Small. It features three key innovations: an efficient channel attention (ECA) module to enhance feature discriminability, an original adaptive multi-scale contrastive regression (AMCR) loss function that extends contrastive learning to multi-dimensional regression for length and weight simultaneously to mitigate data imbalance, and a dependency-graph-based (DepGraph) structured pruning technique that synergistically optimizes model size and performance. On our multi-scene largemouth bass dataset, the pruned ECR-MobileNet-P model comprehensively outperformed 14 mainstream benchmarks. It achieved an R2 of 0.9784 and a root mean square error (RMSE) of 0.4296 cm for length prediction, as well as an R2 of 0.9740 and an RMSE of 0.0202 kg for weight prediction. The model’s parameter count is only 0.52 M, with a computational load of 0.07 giga floating-point operations per second (GFLOPs) and a CPU latency of 10.19 ms, achieving Pareto optimality. This study provides an edge-deployable solution for stress-free biometric monitoring in aquaculture and establishes an innovative methodological paradigm for imbalanced regression and task-oriented model compression.
Open AccessECR-MobileNet: An Imbalanced Largemouth Bass Parameter Prediction Model with Adaptive Contrastive Regression and Dependency-Graph Pruning
The precise, non-destructive monitoring of fish length and weight is a core technology for advancing intelligent aquaculture. However, this field faces dual challenges: traditional contact-based measurements induce stress and yield loss. In addition, existing computer vision methods are hindered by prediction biases from imbalanced data and the deployment bottleneck of balancing high accuracy with model lightweighting. This study aims to overcome these challenges by developing an efficient and robust deep learning framework. We propose ECR-MobileNet, a lightweight framework built on MobileNetV3-Small. It features three key innovations: an efficient channel attention (ECA) module to enhance feature discriminability, an original adaptive multi-scale contrastive regression (AMCR) loss function that extends contrastive learning to multi-dimensional regression for length and weight simultaneously to mitigate data imbalance, and a dependency-graph-based (DepGraph) structured pruning technique that synergistically optimizes model size and performance. On our multi-scene largemouth bass dataset, the pruned ECR-MobileNet-P model comprehensively outperformed 14 mainstream benchmarks. It achieved an R2 of 0.9784 and a root mean square error (RMSE) of 0.4296 cm for length prediction, as well as an R2 of 0.9740 and an RMSE of 0.0202 kg for weight prediction. The model’s parameter count is only 0.52 M, with a computational load of 0.07 giga floating-point operations per second (GFLOPs) and a CPU latency of 10.19 ms, achieving Pareto optimality. This study provides an edge-deployable solution for stress-free biometric monitoring in aquaculture and establishes an innovative methodological paradigm for imbalanced regression and task-oriented model compression.
- Open Access
YOLO-TPS: A Multi-Module Synergistic High-Precision Fish-Disease Detection Model for Complex Aquaculture Environments
Fish are a vital aquatic resource worldwide, and the sustainable development of aquaculture is essential for global food security and economic growth. However, the high incidence of fish diseases in complex aquaculture environments significantly hampers sustainability, and traditional manual diagnosis methods are inefficient and often inaccurate. To address the challenges of small-lesion detection, lesion area size and morphological variation, and background complexity, we propose YOLO-TPS, a high-precision fish-disease detection model based on an improved YOLOv11n architecture. The model integrates a multi-module synergy strategy and a triple-attention mechanism to enhance detection performance. Specifically, the SPPF_TSFA module is introduced into the backbone to fuse spatial, channel, and neuron-level attention for better multi-scale feature extraction of early-stage lesions. A PC_Shuffleblock module incorporating asymmetric pinwheel-shaped convolutions is embedded in the detection head to improve spatial awareness and texture modeling under complex visual conditions. Additionally, a scale-aware dynamic intersection over union (SDIoU) loss function was designed to accommodate changes in the scale and morphology of lesions at different stages of the disease. Experimental results on a dataset comprising 4596 images across six fish-disease categories demonstrate superior performance (mAP0.5: 97.2%, Precision: 97.9%, Recall: 95.1%) compared to the baseline. This study offers a robust, scalable solution for intelligent fish-disease diagnosis and has promising implications for sustainable aquaculture and animal health monitoring.
Open AccessYOLO-TPS: A Multi-Module Synergistic High-Precision Fish-Disease Detection Model for Complex Aquaculture Environments
Fish are a vital aquatic resource worldwide, and the sustainable development of aquaculture is essential for global food security and economic growth. However, the high incidence of fish diseases in complex aquaculture environments significantly hampers sustainability, and traditional manual diagnosis methods are inefficient and often inaccurate. To address the challenges of small-lesion detection, lesion area size and morphological variation, and background complexity, we propose YOLO-TPS, a high-precision fish-disease detection model based on an improved YOLOv11n architecture. The model integrates a multi-module synergy strategy and a triple-attention mechanism to enhance detection performance. Specifically, the SPPF_TSFA module is introduced into the backbone to fuse spatial, channel, and neuron-level attention for better multi-scale feature extraction of early-stage lesions. A PC_Shuffleblock module incorporating asymmetric pinwheel-shaped convolutions is embedded in the detection head to improve spatial awareness and texture modeling under complex visual conditions. Additionally, a scale-aware dynamic intersection over union (SDIoU) loss function was designed to accommodate changes in the scale and morphology of lesions at different stages of the disease. Experimental results on a dataset comprising 4596 images across six fish-disease categories demonstrate superior performance (mAP0.5: 97.2%, Precision: 97.9%, Recall: 95.1%) compared to the baseline. This study offers a robust, scalable solution for intelligent fish-disease diagnosis and has promising implications for sustainable aquaculture and animal health monitoring.
- Open Access
- 9 July 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (14) , 7671
PSCDR-BMPNet: A Point-Supervised Contrastive Deep Regression Network for Point Cloud Biomass Prediction
Accurate assessment of above-ground biomass (AGB) is essential for optimizing crop growth and enhancing agricultural efficiency. However, predicting above-ground biomass (AGB) presents significant challenges. Traditional point cloud networks often struggle with processing crop structures and data characteristics, hindering their ability to predict biomass accurately. To address these limitations, we propose a point-supervised contrastive deep regression method (PSCDR) and a novel network, BMP_Net (BioMixerPoint_Net). The PSCDR method leverages the benefits of deep contrastive regression while accounting for the modal differences between point cloud and 2D image data. By incorporating Chamfer distance to measure point cloud similarity, it improves the model’s adaptability to point cloud features. Experimental results on the SGCBP public dataset show that PSCDR significantly reduces prediction errors compared to seven other point cloud models. Furthermore, the BMP_Net network, which integrates the novel PFMixer module and a point cloud downsampling module, effectively captures the relationship between point cloud structure, density, and biomass. The model achieved test results with RMSE, MAE, and MAPE values of 75.92, 63.19, and 0.115, respectively, outperforming PointMixer by 37.94, 30.07, and 0.079. This method provides an efficient biomass monitoring tool for precision agriculture.
- 9 July 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (14) , 7671
Open AccessPSCDR-BMPNet: A Point-Supervised Contrastive Deep Regression Network for Point Cloud Biomass Prediction
Accurate assessment of above-ground biomass (AGB) is essential for optimizing crop growth and enhancing agricultural efficiency. However, predicting above-ground biomass (AGB) presents significant challenges. Traditional point cloud networks often struggle with processing crop structures and data characteristics, hindering their ability to predict biomass accurately. To address these limitations, we propose a point-supervised contrastive deep regression method (PSCDR) and a novel network, BMP_Net (BioMixerPoint_Net). The PSCDR method leverages the benefits of deep contrastive regression while accounting for the modal differences between point cloud and 2D image data. By incorporating Chamfer distance to measure point cloud similarity, it improves the model’s adaptability to point cloud features. Experimental results on the SGCBP public dataset show that PSCDR significantly reduces prediction errors compared to seven other point cloud models. Furthermore, the BMP_Net network, which integrates the novel PFMixer module and a point cloud downsampling module, effectively captures the relationship between point cloud structure, density, and biomass. The model achieved test results with RMSE, MAE, and MAPE values of 75.92, 63.19, and 0.115, respectively, outperforming PointMixer by 37.94, 30.07, and 0.079. This method provides an efficient biomass monitoring tool for precision agriculture.
- Open Access
YOLO-ALW: An Enhanced High-Precision Model for Chili Maturity Detection
Chili pepper, a widely cultivated and consumed crop, faces challenges in accurately determining maturity due to issues such as occlusion, small target size, and similarity between fruit color and background. This study presents an enhanced YOLOv8n-based object detection model, YOLO-ALW, designed to address these challenges. The model introduces the AKConv (Alterable Kernel Convolution) module in the head section, which adaptively adjusts the convolution kernel shape and size based on the target and scene, improving detection performance under occlusion and dense environments. In the backbone, the SPPF_LSKA (Spatial Pyramid Pooling Fast-Large Separable Kernel Attention) module enhances the integration of multi-scale features, facilitating accurate differentiation of peppers at various maturity stages while maintaining low computational complexity. Additionally, the Wise-IoU (Wise Intersection over Union) loss function optimizes bounding box learning, further improving the detection of peppers in occluded or background-similar scenarios. Experimental results demonstrate that YOLO-ALW achieves a mean average precision (mAP0.5) of 99.1%, with precision and recall rates of 98.3% and 97.8%, respectively, outperforming the original YOLOv8n by 3.4%, 5.1%, and 9.0%, respectively. Grad-CAM feature visualization highlights the model’s improved focus on key fruit features. YOLO-ALW shows significant promise for high-precision chili pepper detection and maturity recognition, offering valuable support for automated harvesting applications.
Open AccessYOLO-ALW: An Enhanced High-Precision Model for Chili Maturity Detection
Chili pepper, a widely cultivated and consumed crop, faces challenges in accurately determining maturity due to issues such as occlusion, small target size, and similarity between fruit color and background. This study presents an enhanced YOLOv8n-based object detection model, YOLO-ALW, designed to address these challenges. The model introduces the AKConv (Alterable Kernel Convolution) module in the head section, which adaptively adjusts the convolution kernel shape and size based on the target and scene, improving detection performance under occlusion and dense environments. In the backbone, the SPPF_LSKA (Spatial Pyramid Pooling Fast-Large Separable Kernel Attention) module enhances the integration of multi-scale features, facilitating accurate differentiation of peppers at various maturity stages while maintaining low computational complexity. Additionally, the Wise-IoU (Wise Intersection over Union) loss function optimizes bounding box learning, further improving the detection of peppers in occluded or background-similar scenarios. Experimental results demonstrate that YOLO-ALW achieves a mean average precision (mAP0.5) of 99.1%, with precision and recall rates of 98.3% and 97.8%, respectively, outperforming the original YOLOv8n by 3.4%, 5.1%, and 9.0%, respectively. Grad-CAM feature visualization highlights the model’s improved focus on key fruit features. YOLO-ALW shows significant promise for high-precision chili pepper detection and maturity recognition, offering valuable support for automated harvesting applications.
- Open Access
- 25 June 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (13) , 5524
YOLO-Chili: An Efficient Lightweight Network Model for Localization of Pepper Picking in Complex Environments
Currently, few deep models are applied to pepper-picking detection, and existing generalized neural networks face issues such as large model parameters, prolonged training times, and low accuracy. To address these challenges, this paper proposes the YOLO-chili target detection algorithm for chili pepper detection. Initially, the classical target detection algorithm YOLOv5 serves as the benchmark model. We introduce an adaptive spatial feature pyramid structure that combines the attention mechanism and the concept of multi-scale prediction to enhance the model’s detection capabilities for occluded and small target peppers. Subsequently, we incorporate a three-channel attention mechanism module to improve the algorithm’s long-distance recognition ability and reduce interference from redundant objects. Finally, we employ a quantized pruning method to reduce model parameters and achieve lightweight processing. Applying this method to our custom chili pepper dataset, we achieve an average precision (AP) value of 93.11% for chili pepper detection, with an accuracy rate of 93.51% and a recall rate of 92.55%. The experimental results demonstrate that YOLO-chili enables accurate and real-time pepper detection in complex orchard environments.
- 25 June 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (13) , 5524
Open AccessYOLO-Chili: An Efficient Lightweight Network Model for Localization of Pepper Picking in Complex Environments
Currently, few deep models are applied to pepper-picking detection, and existing generalized neural networks face issues such as large model parameters, prolonged training times, and low accuracy. To address these challenges, this paper proposes the YOLO-chili target detection algorithm for chili pepper detection. Initially, the classical target detection algorithm YOLOv5 serves as the benchmark model. We introduce an adaptive spatial feature pyramid structure that combines the attention mechanism and the concept of multi-scale prediction to enhance the model’s detection capabilities for occluded and small target peppers. Subsequently, we incorporate a three-channel attention mechanism module to improve the algorithm’s long-distance recognition ability and reduce interference from redundant objects. Finally, we employ a quantized pruning method to reduce model parameters and achieve lightweight processing. Applying this method to our custom chili pepper dataset, we achieve an average precision (AP) value of 93.11% for chili pepper detection, with an accuracy rate of 93.51% and a recall rate of 92.55%. The experimental results demonstrate that YOLO-chili enables accurate and real-time pepper detection in complex orchard environments.
- Open Access
MultiFuseYOLO: Redefining Wine Grape Variety Recognition through Multisource Information Fusion
Based on the current research on the wine grape variety recognition task, it has been found that traditional deep learning models relying only on a single feature (e.g., fruit or leaf) for classification can face great challenges, especially when there is a high degree of similarity between varieties. In order to effectively distinguish these similar varieties, this study proposes a multisource information fusion method, which is centered on the SynthDiscrim algorithm, aiming to achieve a more comprehensive and accurate wine grape variety recognition. First, this study optimizes and improves the YOLOV7 model and proposes a novel target detection and recognition model called WineYOLO-RAFusion, which significantly improves the fruit localization precision and recognition compared with YOLOV5, YOLOX, and YOLOV7, which are traditional deep learning models. Secondly, building upon the WineYOLO-RAFusion model, this study incorporated the method of multisource information fusion into the model, ultimately forming the MultiFuseYOLO model. Experiments demonstrated that MultiFuseYOLO significantly outperformed other commonly used models in terms of precision, recall, and F1 score, reaching 0.854, 0.815, and 0.833, respectively. Moreover, the method improved the precision of the hard to distinguish Chardonnay and Sauvignon Blanc varieties, which increased the precision from 0.512 to 0.813 for Chardonnay and from 0.533 to 0.775 for Sauvignon Blanc. In conclusion, the MultiFuseYOLO model offers a reliable and comprehensive solution to the task of wine grape variety identification, especially in terms of distinguishing visually similar varieties and realizing high-precision identifications.
Open AccessMultiFuseYOLO: Redefining Wine Grape Variety Recognition through Multisource Information Fusion
Based on the current research on the wine grape variety recognition task, it has been found that traditional deep learning models relying only on a single feature (e.g., fruit or leaf) for classification can face great challenges, especially when there is a high degree of similarity between varieties. In order to effectively distinguish these similar varieties, this study proposes a multisource information fusion method, which is centered on the SynthDiscrim algorithm, aiming to achieve a more comprehensive and accurate wine grape variety recognition. First, this study optimizes and improves the YOLOV7 model and proposes a novel target detection and recognition model called WineYOLO-RAFusion, which significantly improves the fruit localization precision and recognition compared with YOLOV5, YOLOX, and YOLOV7, which are traditional deep learning models. Secondly, building upon the WineYOLO-RAFusion model, this study incorporated the method of multisource information fusion into the model, ultimately forming the MultiFuseYOLO model. Experiments demonstrated that MultiFuseYOLO significantly outperformed other commonly used models in terms of precision, recall, and F1 score, reaching 0.854, 0.815, and 0.833, respectively. Moreover, the method improved the precision of the hard to distinguish Chardonnay and Sauvignon Blanc varieties, which increased the precision from 0.512 to 0.813 for Chardonnay and from 0.533 to 0.775 for Sauvignon Blanc. In conclusion, the MultiFuseYOLO model offers a reliable and comprehensive solution to the task of wine grape variety identification, especially in terms of distinguishing visually similar varieties and realizing high-precision identifications.
- Open Access
- 29 April 2024
- Published by MDPI AG in Preprints.org
Yolo-Chili: An Efficient Lightweight Network Model for Localization of Pepper Picking in Complex Environments
Currently there are fewer depth models applied to pepper picking detection, while the existing generalized neural networks have problems such as large model parameters, long training time, and low model accuracy.In order to solve the above problems, this paper proposes a Yolo-chili target detection algorithm for chili pepper detection. First, the classical target detection algorithm yolov5 is used as a benchmark model, and an adaptive spatial feature pyramid structure combining the attention mechanism and the idea of multi-scale prediction is introduced to improve the model's detection effect on occluded peppers and small target peppers. Secondly, a three-channel attention mechanism module is introduced to improve the algorithm's long-distance recognition ability and reduce the interference of redundant testers. Finally, the quantized pruning method is used to reduce the model parameters and realize the lightweight processing of the model. Applying the method to the homemade chili pepper dataset, the AP value of chili pepper reaches 93.11%; the accuracy rate is 93.51% and the recall rate is 92.55%.The experimental results show that yolo-chili is able to achieve accurate and real-time pepper detection under complex orchards.
- 29 April 2024
- Published by MDPI AG in Preprints.org
Open AccessYolo-Chili: An Efficient Lightweight Network Model for Localization of Pepper Picking in Complex Environments
Currently there are fewer depth models applied to pepper picking detection, while the existing generalized neural networks have problems such as large model parameters, long training time, and low model accuracy.In order to solve the above problems, this paper proposes a Yolo-chili target detection algorithm for chili pepper detection. First, the classical target detection algorithm yolov5 is used as a benchmark model, and an adaptive spatial feature pyramid structure combining the attention mechanism and the idea of multi-scale prediction is introduced to improve the model's detection effect on occluded peppers and small target peppers. Secondly, a three-channel attention mechanism module is introduced to improve the algorithm's long-distance recognition ability and reduce the interference of redundant testers. Finally, the quantized pruning method is used to reduce the model parameters and realize the lightweight processing of the model. Applying the method to the homemade chili pepper dataset, the AP value of chili pepper reaches 93.11%; the accuracy rate is 93.51% and the recall rate is 92.55%.The experimental results show that yolo-chili is able to achieve accurate and real-time pepper detection under complex orchards.
- Open Access
Enhancing Aboveground Biomass Prediction through Integration of the SCDR Paradigm into the U-Like Hierarchical Residual Fusion Model
Deep learning methodologies employed for biomass prediction often neglect the intricate relationships between labels and samples, resulting in suboptimal predictive performance. This paper introduces an advanced supervised contrastive learning technique, termed Improved Supervised Contrastive Deep Regression (SCDR), which is adept at effectively capturing the nuanced relationships between samples and labels in the feature space, thereby mitigating this limitation. Simultaneously, we propose the U-like Hierarchical Residual Fusion Network (BioUMixer), a bespoke biomass prediction network tailored for image data. BioUMixer enhances feature extraction from biomass image data, facilitating information exchange and fusion while considering both global and local features within the images. The efficacy of the proposed method is validated on the Pepper_Biomass dataset, which encompasses over 600 original images paired with corresponding biomass labels. The results demonstrate a noteworthy enhancement in deep regression tasks, as evidenced by performance metrics on the Pepper_Biomass dataset, including RMSE = 252.18, MAE = 201.98, and MAPE = 0.107. Additionally, assessment on the publicly accessible GrassClover dataset yields metrics of RMSE = 47.92, MAE = 31.74, and MAPE = 0.192. This study not only introduces a novel approach but also provides compelling empirical evidence supporting the digitization and precision improvement of agricultural technology. The research outcomes align closely with the identified problem and research statement, underscoring the significance of the proposed methodologies in advancing the field of biomass prediction through state-of-the-art deep learning techniques.
Open AccessEnhancing Aboveground Biomass Prediction through Integration of the SCDR Paradigm into the U-Like Hierarchical Residual Fusion Model
Deep learning methodologies employed for biomass prediction often neglect the intricate relationships between labels and samples, resulting in suboptimal predictive performance. This paper introduces an advanced supervised contrastive learning technique, termed Improved Supervised Contrastive Deep Regression (SCDR), which is adept at effectively capturing the nuanced relationships between samples and labels in the feature space, thereby mitigating this limitation. Simultaneously, we propose the U-like Hierarchical Residual Fusion Network (BioUMixer), a bespoke biomass prediction network tailored for image data. BioUMixer enhances feature extraction from biomass image data, facilitating information exchange and fusion while considering both global and local features within the images. The efficacy of the proposed method is validated on the Pepper_Biomass dataset, which encompasses over 600 original images paired with corresponding biomass labels. The results demonstrate a noteworthy enhancement in deep regression tasks, as evidenced by performance metrics on the Pepper_Biomass dataset, including RMSE = 252.18, MAE = 201.98, and MAPE = 0.107. Additionally, assessment on the publicly accessible GrassClover dataset yields metrics of RMSE = 47.92, MAE = 31.74, and MAPE = 0.192. This study not only introduces a novel approach but also provides compelling empirical evidence supporting the digitization and precision improvement of agricultural technology. The research outcomes align closely with the identified problem and research statement, underscoring the significance of the proposed methodologies in advancing the field of biomass prediction through state-of-the-art deep learning techniques.