Improved YOLOv8n Orchard Tomato Target Detection Algorithm

doi:10.3778/j.issn.1002-8331.2406-0132

Abstract

Abstract: In the environment of natural orchards, tomatoes with different growth cycles have various postures, are susceptible to the influence of light and green leaf background, resulting in not obvious image features, and tomato fruits are prone to cluster densely and branches and leaves are blocked by vines during growth, which often causes problems such as missed detection and false detection. A method for picking and detecting tomato growth cycles based on improved YOLOv8n is proposed. Firstly, super-resolution adaptive attention module (SPAAM) is designed in backbone to effectively improve feature image resolution and solve the problem of inadequate feature extraction of small target tomatoes. Combining coordinate attention (CA) can improve the ability of extracting key position information. Secondly, C2F-DCF is designed to replace the original C2f, which is used to adapt the tomato attitude deformation characteristics, improve the modeling ability of the spatial layout of the deformed object, and improve the calculation efficiency. GSCHead is designed to reduce the number of head parameters, and four-fold subsampling branches are added to improve the constraint effect on small target tomatoes. Finally, Wise-IoU loss function is introduced to improve the generalization performance of the model trained on different quality images. The accuracy of the improved algorithm on the test set reaches 93.9%, which is 1.9?percentage points higher than that of the original model, and the number of parameters is reduced by 0.18×106, which effectively improves the missed detection rate of occlusion and the detection performance of small target tomatoes. At the same time, the detection speed reaches 139 FPS, which can be easily deployed to the terminal to complete real-time detection.

Key words: tomato detection, YOLOv8n, super-resolution adaptive attention module (SPAAM), C2f-DCF, GSCHead, loss function

摘要： 针对自然果园环境下，不同生长周期的番茄姿态多变，易受光线、绿色叶片背景影响导致图像特征不明显，番茄果实生长时易出现扎堆密集以及枝叶藤蔓遮挡等情况，时常造成漏检、误检等问题，提出了一种基于改进YOLOv8n的番茄生长周期采摘检测方法。在Backbone中设计超分辨率自适应注意力模块（super-resolution adaptive attention module，SPAAM），有效提升特征图像分辨率，改善小目标番茄特征提取不充分的问题，并结合坐标注意力机制（coordinate attention，CA）提高关键位置信息提取能力；设计C2f-DCF替换原有C2f，用于自适应番茄姿态形变特征，提高对形变物体空间布局的建模能力，同时提升计算效率；设计GSCHead降低头部参数量，并且添加四倍下采样分支提高对小目标番茄的约束效果；引入Wise-IoU损失函数，提升模型在不同质量图像上训练的泛化性能。改进后的算法在测试集上精确率达到93.9%，相较于原模型提升1.9个百分点，参数量降低0.18×106，有效改善了遮挡情况的漏检率和小目标番茄的检测性能，同时检测速度达到139 FPS，可以便捷地部署到终端完成实时检测。

关键词: 番茄检测, YOLOv8n, 超分辨率自适应注意力模块（SPAAM）, C2f-DCF, GSCHead, 损失函数

YANG Guoliang, SHENG Yangyang, HONG Xinfang, ZHANG Jiaqi. Improved YOLOv8n Orchard Tomato Target Detection Algorithm[J]. Computer Engineering and Applications, 2024, 60(23): 238-248.

杨国亮, 盛杨杨, 洪鑫芳, 张佳琦. 改进YOLOv8n的果园番茄目标检测算法[J]. 计算机工程与应用, 2024, 60(23): 238-248.

References

[1] 沈衡, 王琳, 李骞, 等. 番茄风味和功能性成分研究进展[J]. 园艺学报, 2024, 51(2): 423-438.
SHEN H, WANG L, LI Q, et al. Analyzing flavor and functional components in tomatoes：a review[J]. Acta Horticulturae Sinica, 2024, 51(2): 423-438.
[2] 陈林蔚. 区块链技术在智慧农业领域中的应用[J]. 江苏农业学报, 2023, 39(6): 1358-1365.
CHEN L W. The application of blockchain technology in the field of smart agriculture[J]. Jiangsu Journal of Agricultural Sciences, 2023, 39(6): 1358-1365.
[3] 鲍秀兰, 马志涛, 马萧杰, 等. 丘陵果园自然环境下柑橘采摘机器人设计与试验[J]. 农业机械学报, 2024, 55(4): 124-135.
BAO X L, MA Z T, MA X J, et al. Design and experiment of citrus picking robot in hilly orchard natural environment[J]. Transactions of the Chinese Society for Agricultural Machinery, 2024, 55(4): 124-135.
[4] 张永宏, 李宇超, 董天天, 等. 非结构化环境下番茄采摘机器人目标识别与检测[J]. 中国农机化学报, 2024, 45(4): 205-213.
ZHANG Y H, LI Y C, DONG T T, et al. Target identification and detection for tomato harvesting robot in unstructured environments[J]. Journal of Chinese Agricultural Mechanization, 2024, 45(4): 205-213.
[5] GIRSHICK R. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision, 2015: 1440-1448.
[6] FU L, FENG Y, MAJEED Y, et al. Kiwifruit detection in field images using Faster R-CNN with ZFNet[J]. IFAC-PapersOnLine, 2018, 51(17): 45-50.
[7] 王梁, 侯义锋, 贺杰. 基于Mask-RCNN的自然场景下油茶果目标识别与检测[J]. 中国农机化学报, 2022, 43(12): 148-154.
WANG L, HOU Y F, HE J. Target recognition and detection of Camellia oleifera fruit in natural scene based on Mask-RCNN[J]. Journal of Chinese Agricultural Mechanization, 2022, 43(12): 148-154.
[8] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 779-788.
[9] LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision, 2016: 21-37.
[10] 高新阳, 魏晟, 温志庆, 等. 改进YOLOv5轻量级网络的柑橘检测方法[J]. 计算机工程与应用, 2023, 59(11): 212-221.
GAO X Y, WEI S, WEN Z Q, et al. Citrus detection method based on improved YOLOv5 lightweight network[J]. Computer Engineering and Applications, 2023, 59(11): 212-221.
[11] 赵鹏飞, 钱孟波, 周凯琪, 等. 改进YOLOv7-Tiny农田环境下甜椒果实检测[J]. 计算机工程与应用, 2023, 59(15): 329-340.
ZHAO P F, QIAN M B, ZHOU K Q, et al. Improvement of sweet pepper fruit detection in YOLOv7-tiny farming environment[J]. Computer Engineering and Applications, 2023, 59(15): 329-340.
[12] WANG J, CHEN Y, DONG Z, et al. Improved YOLOv5 network for real-time multi-scale traffic sign detection[J]. Neural Computing and Applications, 2023, 35(10): 7853-7865.
[13] ZHANG Y, LI K, LI K, et al. Image super-resolution using very deep residual channel attention networks[C]//Proceedings of the European Conference on Computer Vision, 2018: 286-301.
[14] HOU Q, ZHOU D, FENG J. Coordinate attention for efficient mobile network design[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 13713-13722.
[15] DAI J, QI H, XIONG Y, et al. Deformable convolutional networks[C]//Proceedings of the IEEE International Conference on Computer Vision, 2017: 764-773.
[16] CHEN J, KAO S, HE H, et al. Run, don’t walk: chasing higher FLOPS for faster neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 12021-12031.
[17] LI H, LI J, WEI H, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[J]. arXiv:2206.02424, 2022.
[18] TONG Z, CHEN Y, XU Z, et al. Wise-IoU: bounding box regression loss with dynamic focusing mechanism[J]. arXiv:2301.10051, 2023.
[19] 胡奕帆, 赵贤林, 李佩娟, 等. 基于改进YOLOv5的自然环境下番茄果实检测[J]. 中国农机化学报, 2023, 44(10): 231-237.
HU Y F, ZHAO X L, LI P J, et al. Tomato fruit detection in natural environment based on improved YOLOv5[J]. Journal of Chinese Agricultural Mechanization, 2023, 44(10): 231-237.
[20] 张俊宁, 毕泽洋, 闫英, 等. 基于注意力机制与改进YOLO的温室番茄快速识别[J]. 农业机械学报, 2023, 54(5): 236-243.
ZHANG J N, BI Z Y, YAN Y, et al. Fast recognition of greenhouse tomato targets based on attention mechanism and improved YOLO[J]. Transactions of the Chinese Society for Agricultural Machinery, 2023, 54(5): 236-243.
[21] 李慧琴, 宋赵铭, 刘存祥, 等. 基于YOLOv8n的番茄果实检测模型改进[J/OL]. 河南农业大学学报: 1-14[2024-06-27]. https://doi.org/10.16445/j.cnki.1000-2340.20240511.002.
LI H Q, SONG Z M, LIU C X, et al. Improvement of tomato fruit detection model based on YOLOv8n[J/OL]. Journal of Hennan Agricultural University: 1-14[2024-06-27]. https://doi.org/10.16445/j.cnki.1000-2340.20240511.002.