自然环境下一种改进的 YOLOv11-ISAB 大樱桃裂果检测模型设计

Design of an Improved YOLOv11-ISAB Cherry Fruit Cracking Detection Model in Natural Environments

摘要

近年来，我国樱桃产业规模持续扩大，大连等主产区产量逐年攀升，但机械化采摘分拣仍面临技术瓶颈。为满足樱桃采摘机器人在复杂果园场景下对裂果的精准识别需求，提升采摘分拣效率，降低农户劳动强度，本文从优化YOLOv11检测性能出发，提出了一种改进的YOLOv11-ISAB目标检测方法。首先，引入Inner-SIoU损失函数，通过优化边界框回归的形状、方向及距离约束，提升裂果定位精度；其次，设计了RFCBEMConv卷积，并替换网络中的标准卷积，实现对高维中间特征执行多尺度注意力提炼，强化裂果细微纹理与全局形态的特征表达；最后，提出了一种并联双注意力结构BDA，通过门控机制动态融合方向感知与长距离语义关联信息，利用该BDA模块重构原始C3K2模块中的Bottleneck部分，形成改进的C3K2_BDA模块，以精准区分裂果特征与背景干扰。在自采樱桃裂果数据集上的实验表明：相较于原始YOLOv11，所提方法的mAP@0.5提升1.6个百分点，裂果检测精确率提升5.4个百分点，mAP@0.5~0.95提升3.8个百分点，模型对复杂光照、遮挡场景的鲁棒性显著增强。该研究为农业樱桃采摘分拣机器人提供了更高效的裂果检测方法，显著提升了裂果识别与定位精度，具备重要的工程应用价值，尤其适用于大连等地大面积樱桃产区的智能机器人采摘与分拣协同作业需求。

Abstract

In recent years, China's cherry industry has significantly expanded, with yields in major producing regions like Dalian increasing year by year. However, the mechanized harvesting and sorting processes still face technical bottlenecks. To meet the demand for accurate detection of cracked fruits by cherry-harvesting robots in complex orchard environments, improve harvesting and sorting efficiency, and reduce farmers' labor intensity, this paper proposes a YOLOv11-ISAB object detection method, based on optimizing the detection performance of YOLOv11. First, the Inner-SIoU loss function is introduced to improve the localization accuracy of cracked cherries by optimizing shape, orientation, and distance constraints in bounding box regression. Second, a RFCBEMConv(Receptive Field and Channel Boosted Efficient Module Convolution) is designed to replace the standard convolution in the network, enabling multi-scale attention refinement of high-dimensional intermediate features and enhancing the feature representation of fine textures and global morphology of cracked cherries. Finally, a parallel Bottleneck with Dual Attention (BDA)structure is proposed, which dynamically fuses orientation-aware and long-range semantic correlation information through a gating mechanism. Using this BDA module, the Bottleneck part in the original C3K2 module is reconstructed to form an improved C3K2_BDA module, enabling accurate differentiation of cracked cherry features from background interference. Experiments conducted on a self-built dataset of cracked cherries demonstrate that compared to the original YOLOv11, the proposed method achieves a 1.6 percentage point increase in mAP@0.5,a 5.4 percentage point gain in precision,and a 3.8 percentage point improvement in mAP@0.5~0.95 over the baseline YOLOv11. The model also exhibits significantly enhanced robustness in complex lighting and occlusion. This research provides a more efficient detection solution for cracked fruit identification in agricultural cherry-picking robots, markedly improving detection and localization accuracy. It holds substantial engineering application value, particularly in facilitating intelligent robotic harvesting and collaborative sorting operations in large-scale cherry production areas such as Dalian.

关键词

大樱桃裂果 / 目标检测 / YOLOv11-ISAB / 自然环境

Key words

Large cherry cracked fruit / object detection / YOLOv11-ISAB / natural environment

参考文献

[1] 苗荣慧, 李志伟, 武锦龙. 基于改进YOLO v7的轻量化樱桃番茄成熟度检测[J]. 农业机械学报, 2023, 54(10): 225-233.

[2] 张俊宁, 闫英, 胡欢, 等. 基于Sim-YOLO的复杂背景下樱桃番茄多目标检测[J]. 中国农机化学报, 2025, 46(03): 238-245.

[3] 张志远, 罗铭毅, 郭树欣, 等. 基于改进YOLO v5的自然环境下樱桃果实识别方法[J]. 农业机械学报, 2022, 53(S1): 232-240.

[4] 李茂, 肖洋轶, 宗望远. 基于改进YOLO 11的多尺度板栗果实识别方法[J]. 农业机械学报, 2025, 56(05): 443-454.

[5] 李晋超, 朱晓军. 改进 YOLOv5s 的葡萄采摘机器人果实检测[J]. 机械设计与制造, 2025, 45(05): 1-8.

[6] 文瑞. 基于YOLOX-CA 算法的玉米花丝检测模型研究与应用[D].重庆:重庆师范大学, 2023.

[7] 侯依廷, 饶元, 宋贺, 等. 复杂大田场景下基于改进YOLOv8的小麦幼苗期叶片数快速检测方法[J]. 智慧农业(中英文), 2024, 6(04): 128-137.

[8] Wang L, Wang S, Wang B, et al. Jujube-YOLO: A precise jujube fruit recognition model in unstructured environments[J]. Expert Systems With Applications, 2025, 291: 128530-128540.

[9] Cai Z, Zhang Y, Li J, et al. Synchronous detection of internal and external defects of citrus by structured-illumination reflectance imaging coupling with improved YOLOv7[J]. Postharvest Biology and Technology, 2025, 227: 113576-113586.

[10] Yi F B, Jun Z Y, Shu Q Y, et al. An improved YOLO algorithm for detecting flowers and fruits on strawberry seedlings[J]. Biosystems Engineering, 2024, 237: 1-12.

[11] 徐胜勇, 易同舟, 秦子轶, 等. 基于单视角 RGBD 图像的柑橘果实三维重建与表型检测方法[J]. 农业机械学报, 2025, 56(03): 80-90.

[12] Li H, Mo Y, Chen J, et al. Accurate Orah fruit detection method using lightweight improved YOLOv8n model verified by optimized deployment on edge device[J]. Artificial Intelligence in Agriculture, 2025, 15(4): 707-723.

[13] Villacrés J F, Auat C F . Detection and characterization of cherries: A deep learning usability case study in chile[J]. Agronomy, 2020, 10(6): 835-845.

[14] 张永飞, 裴悦琨, 姜艳超, 等. 基于深度卷积神经网络的樱桃分级检测[J]. 食品研究与开发, 2021, 42(14): 138-144.

[15] 张亚亚. 基于深度学习的樱桃图像检测研究[D].天水:天水师范学院, 2023.

[16] 冯良枫. 基于深度学习的樱桃番茄虫害区域和生长状态识别算法研究[D].广州:华南理工大学, 2023.

[17] 帖军, 赵捷, 郑禄, 等. 改进 YOLOv5 模型在自然环境下柑橘识别的应用[J]. 中国农业科技导报, 2024, 26(07): 111-120.

[18] 李彬, 李生林. 改进YOLOv11n 的无人机小目标检测算法[J]. 计算机工程与应用, 2025, 61(08): 1-11.

[19] Jegham N, Koh C Y, Abdelatti M, et al. Evaluating the evolution of yolo(you only look once)models: A comprehensive benchmark study of yolo11 and its predecessors[EB]. (2024-10-01) [2025-04-21]. https://arxiv.org/abs/2411.00201.

[20] 李成轩, 赵春江, 张驰, 等. 基于AIC-YOLOv11n 模型的砀山梨多表面缺陷检测方法[J]. 农业工程学报, 2025, 41(07): 1-9.

[21] 秦玮健, 康峰, 王亚雄, 等. 基于改进YOLOv8 算法的枸杞果实目标检测[J]. 机械工程与自动化, 2024(06): 24-26+ 30.

[22] Badgujar C M, Poulose A, Gan H . Agricultural object detection with You Only Look Once (YOLO) Algorithm: A bibliometric and systematic literature review[J]. Computers and Electronics in Agriculture, 2024, 223: 109090-109100.

[23] Zhang M, Ye S, Zhao S, et al. Pear object detection in complex orchard environment based on improved YOLO11[J]. Symmetry, 2025, 17(2): 255-265.

[24] Pranav Adarsh , Pratibha Rathi , Manoj Kumat . YOLO v3-Tiny: Object Detection and Recognition using one stage improved model[C]. 2020 6th International Conference on Advanced Computing and Communication Systems: 6th International Conference on Advanced Computing and Communication Systems (ICACCS 2020), 6-7 March 2020, Coimbatore, India.: Institute of Electrical and Electronics Engineers, 2020: 687-694.

[25] Li C, Li L, Jiang H, et al. YOLOv6: A Single-stage object detection framework for industrial applications[J]. arXiv Preprint arXiv:2209.02976, 2022.

[26] Chen H, Zhou G, Jiang H . Student behavior detection in the classroom based on improved YOLOv8[J]. Sensors, 2023, 23(15): 8385-8395.

[27] Gharat K, Jogi H, Gode K, et al. Enhanced detection of maize leaf blight in dynamic FieldConditions using modified YOLOv9[C]. 2024 IEEE Space, Aerospace and Defence Conference (SPACE).IEEE, 2024: 140-143.

[28] Wang A, Chen H, Liu L, et al. Yolov10: Real-time end-to-end object detection[J]. Advances in Neural Information Processing Systems, 2024, 37: 107984-108011.

[29] Alif Mujadded Al Rabbani , Muhammad Hussain . YOLOv12: A breakdown of the key architectural features[EB].(2025-02)[ 2025-05-01]. https://arxiv.org/abs/2502.14740.

[30] Ren S, He K, Girshick R, et al. Faster R-CNN:Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(6): 1137-1149.

基金资助

辽宁省科技计划重点项目(2022020655-JH1/109)

大连市科技创新基金项目(2022JJ12SN052)