湖北农业科学 ›› 2023, Vol. 62 ›› Issue (11): 183-190.doi: 10.14088/j.cnki.issn0439-8114.2023.11.032

• 农业工程 • 上一篇    下一篇

基于改进Faster R-CNN模型的草莓果实识别算法

李佳俊, 朱子峰, 刘洪鑫, 苏昱荣, 温传闻, 张原升, 张慧敏, 邓立苗   

  1. 青岛农业大学理学与信息科学学院,山东 青岛 266109
  • 收稿日期:2022-04-12 出版日期:2023-11-25 发布日期:2023-12-25
  • 通讯作者: 邓立苗(1978-),女,山东沂水人,副教授,博士,主要从事机器学习方面的研究,(电话)13626484274(电子信箱)denglm68@163.com。
  • 作者简介:李佳俊(2000-),男,山东新泰人,在读硕士研究生,研究方向为农业作物目标检测,(电话)17664092106(电子信箱)jerryleereal@163.com。
  • 基金资助:
    国家自然科学基金项目(31872849; 32073029),山东省高等学校青创人才引育计划项目(008/1622001); 青岛农业大学高层次人才基金项目(1119036)

Strawberry fruit recognition algorithm based on improved Faster R-CNN model

LI Jia-jun, ZHU Zi-feng, LIU Hong-xin, SU Yu-rong, WEN Chuan-wen, ZHANG Yuan-sheng, ZHANG Hui-min, DENG Li-miao   

  1. School of Science and Information, Qingdao Agricultural University, Qingdao 266109, Shandong,China
  • Received:2022-04-12 Online:2023-11-25 Published:2023-12-25

摘要: 针对Faster R-CNN模型对自然状态下草莓(Fragaria ananassa Duch.)识别准确率不高的问题,以地垄种植草莓的实拍图片为数据源,采用改进RPN结构和更换主干特征提取网络的方法对Faster R-CNN模型进行了改进。结果表明,改进Faster R-CNN模型识别成熟草莓平均精度(AP)为0.893 0,识别未成熟草莓平均精度(AP)为0.820 7,草莓识别准确率达到较高水平,解决了未成熟草莓识别困难的问题。同时,为了检验模型的自动计数性能,依据模型的识别结果建立了自动计数与人工计数的线性回归,成熟草莓、未成熟草莓的相关系数分别为0.973 7、0.944 7,自动计数与人工计数拥有较高的相关性,表明改进Faster R-CNN模型具有较高的识别性能与计数能力。

关键词: 草莓(Fragaria ananassa Duch.), 识别, Faster R-CNN模型, ResNet50

Abstract: In response to the problem of low recognition accuracy of the Faster R-CNN model for natural strawberries (Fragaria ananassa Duch.), the Faster R-CNN model was improved by improving the RPN structure and replacing the backbone feature extraction network using live images of strawberries planted on ridges as the data source.The results showed that the improved Faster R-CNN model had an average precision (AP) of 0.893 0 when identifying mature strawberries and 0.820 7 when identifying immature strawberries. The accuracy of strawberry recognition reached a high level, solving the problem of difficulty in identifying immature strawberries.Meanwhile, in order to test the automatic counting performance of the model, a linear regression between automatic counting and manual counting was established based on the recognition results of the model. The correlation coefficients of mature and immature strawberries were 0.973 7 and 0.944 7, respectively. The high correlation between automatic counting and manual counting indicated that the improved Faster R-CNN model had high recognition performance and counting ability.

Key words: strawberry (Fragaria ananassa Duch.), identification, Faster R-CNN model, ResNet50

中图分类号: