Detection of Pear Inflorescence Based on Improved Ghost-YOLOv5s-BiFPN Algorithm

doi:10.12133/j.smartag.SA202207006

Abstract

Abstract:

Mechanized and intelligent flower thinning is a high-speed flower thinning method nowadays. The classification and detection of flowers and flower buds are the basic requirements to ensure the normal operation of the flower thinning machine. Aiming at the problems of pear inflorescence detection and classification in the current intelligent production of pear orchards, a Y-shaped shed pear orchard inflorescence recognition algorithm Ghost-YOLOv5s-BiFPN based on improved YOLOv5s was proposed in this research. The detection model was obtained by labeling and expanding the pear tree bud and flower images collected in the field and sending them to the algorithm for training. The Ghost-YOLOv5s-BiFPN algorithm used the weighted bidirectional feature pyramid network to replace the original path aggregation network structure, and effectively fuse the features of different sizes. At the same time, ghost module was used to replace the traditional convolution, so as to reduce the amount of model parameters and improve the operation efficiency of the equipment without reducing the accuracy. The field experiment results showed that the detection accuracy of the Ghost-YOLOv5s-BiFPN algorithm for the bud and flower in the pear inflorescence were 93.21% and 89.43%, respectively, with an average accuracy of 91.32%, and the detection time of a single image was 29 ms. Compared with the original YOLOv5s algorithm, the detection accuracy was improved by 4.18%, and the detection time and model parameters were reduced by 9 ms and 46.63% respectively. Compared with the original YOLOV5s network, the mAP and recall rate were improved by 4.2% and 2.7%, respectively; the number of parameters, model size and floating point operations were reduced by 46.6%, 44.4% and 47.5% respectively, and the average detection time was shortened by 9 ms. With Ghost convolution and BIFPN adding model, the detection accuracy has been improved to a certain extent, and the model has been greatly lightweight, effectively improving the detect efficiency. From the thermodynamic diagram results, it can be seen that BIFPN structure effectively enhances the representation ability of features, making the model more effective in focusing on the corresponding features of the target. The results showed that the algorithm can meet the requirements of accurate identification and classification of pear buds and flowers, and provide technical support for the follow-up pear garden to achieve intelligent flower thinning.

Key words: pear flower, intelligent recognition, YOLOv5s, BiFPN, lightweight model

CLC Number:

TP391.41

XIA Ye, LEI Xiaohui, QI Yannan, XU Tao, YUAN Quanchun, PAN Jian, JIANG Saike, LYU Xiaolan. Detection of Pear Inflorescence Based on Improved Ghost-YOLOv5s-BiFPN Algorithm[J]. Smart Agriculture, 2022, 4(3): 108-119.

Figures/Tables 12

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Fig. 7

Fig. 8

Fig. 9

Fig. 10

Table1

Fig. 11

References 23

1	张伏, 陈自均, 鲍若飞, 等. 基于改进型YOLOv4-LITE轻量级神经网络的密集圣女果识别[J]. 农业工程学报, 2021, 37(16): 270-278.
	ZHANG F, CHEN Z, BAO R, et al. Recognition of dense cherry tomatoes based on improved YOLOv4-LITE lightweight neural network[J]. Transactions of the CSAE, 2021, 37(16): 270-278.
2	刘天真, 滕桂法, 苑迎春, 等. 基于改进YOLO v3的自然场景下冬枣果实识别方法[J]. 农业机械学报, 2021, 52(5): 17-25.
	LIU T, TENG G, YUAN Y, et al. Winter jujube fruit recognition method based on improved YOLOv3 under natural scene[J]. Transactions of the CSAM, 2021, 52(5): 17-25.
3	KANG H, CHEN C. Fruit detection. segmentation and 3D visualisation of environments in apple orchards[J]. Computers and Electronics in Agriculture, 2020, 171: ID 105302.
4	WANG Y, LYU J, XU L, et al. A segmentation method for waxberry image under orchard environment[J]. Scientia Horticulturae, 2020, 266: ID 109309.
5	杜文圣, 王春颖, 朱衍俊, 等. 采用改进Mask R-CNN算法定位鲜食葡萄疏花夹持点[J]. 农业工程学报, 2022, 38(1): 169-177.
	DU W, WANG C, ZHU Y, et al. Fruit stem clamping points location for table grape thinning using improved mask R-CNN[J]. Transactions of the CSAE, 2022, 38(1): 169-177.
6	陈新, 伍萍辉, 祖绍颖, 等. 基于改进SSD轻量化神经网络的番茄疏花疏果农事识别方法[J]. 中国瓜菜, 2021, 34(9): 38-44.
	CHEN X, WU P, ZU S, et al. Study on identification method of thinning flower and fruit of tomato based on improved SSD lightweight neural network[J]. China Cucurbits and Vegetables, 2021, 34(9): 38-44.
7	龙洁花, 郭文忠, 林森, 等. 改进YOLOv4的温室环境下草莓生育期识别方法[J]. 智慧农业(中英文), 2021, 3(4): 99-110.
	LONG J, GUO W, LIN S, et al. Strawberry growth period recognition method under greenhouse environment based on improved YOLOv4[J]. Smart Agriculture, 2021, 3(4): 99-110.
8	WU D, LYU S, JIANG M, et al. Using channel pruning-based YOLO v4 deep learning algorithm for the real-time and accurate detection of apple flowers in natural environments[J]. Computers and Electronics in Agriculture, 2020, 178: ID 105742.
9	FARJON G, KRIKEB O, HILLEL A, et al. Detection and counting of flowers on apple trees for better chemical thinning decisions[J]. Precision Agriculture, 2020, 21(3): 503-521.
10	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// The IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, New York, USA: IEEE, 2014: 580-587.
11	LIU W, ANGUELOV D, ERHAN D, et al. Ssd: Single shot multibox detector[C]// European Conference on Computer Vision. Berlin, German: Springer, 2016: 21-37.
12	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]// The IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, New York, USA: IEEE, 2016: 779-788.
13	SHAFIEE M J, CHYWL B, LI F, et al. Fast YOLO: A fast you only look once system for real-time embedded object detection in video[J/OL]. arXiv: , 2017.
14	REDMON J, FARHADI A. Yolov3: An incremental improvement[J/OL]. arXiv: , 2018.
15	BOCHKOVSKIY A, WANG C Y, LIAO H Y M. Yolov4: Optimal speed and accuracy of object detection[J/OL]. arXiv: , 2020.
16	HAN K, WANG Y, TIAN Q, et al. Ghostnet: More features from cheap operations[C]// The IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, New York, USA: IEEE, 2020: 1580-1589.
17	HOWARD A G, ZHU M, CHEN B, et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications[J/OL]. arXiv: , 2017.
18	SANDLER M, HOWARD A, ZHU M, et al. MobileNetV2: Inverted Residuals and Linear Bottlenecks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, New York, USA: IEEE, 2018.
19	HOWARD A, SANDLER M, CHU G, et al. Searching for mobilenetv3[C]// The IEEE/CVF International Conference on Computer Vision. Piscataway, New York, USA: IEEE, 2019: 1314-1324.
20	ZHANG X, ZHOU X, LIN M, et al. Shufflenet: An extremely efficient convolutional neural network for mobile devices[C]// The IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, New York, USA: IEEE, 2018: 6848-6856.
21	MA N, ZHANG X, ZHENG H T, et al. Shufflenet v2: Practical guidelines for efficient cnn architecture design[C]// The European Conference on COMPUTER VISIon (ECCV). Piscataway, New York, USA: IEEE, 2018: 116-131.
22	LIU S, QI L, QIN H, et al. Path aggregation network for instance segmentation[C]// The IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, New York, USA: IEEE, 2018: 8759-8768.
23	TAN M, PANG R, LE Q V. Efficientdet: Scalable and efficient object detection[C]// The IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, New York, USA: IEEE, 2020: 10781-10790.

算法	mAP/％	召回率/％	F₁得分/％	参数量	GFLOPs	平均检测时间/ms	模型大小/M
YOLOv5s	87.1	87.2	88.0	7,015,519	15.8	38	13.70
YOLOv5s-BiFPN	92.2	91.4	91.8	7,101,064	16.0	41	14.10
Ghost-YOLOv5s	86.2	86.5	86.5	3,678,423	8.1	27	7.49
Ghost-YOLOv5s-BiFPN	91.3	89.9	91.2	3,743,968	8.3	29	7.62