卷积神经网络
深度学习
温室
跳跃式监视
最小边界框
机器学习
人工神经网络
目标检测
计算机科学
班级(哲学)
模式识别(心理学)
人工智能
园艺
生物
图像(数学)
作者
Harshana Habaragamuwa,Yuichi Ogawa,Tetsuhito Suzuki,Tomoo Shiigi,Masanori Ono,Naoshi Kondo
出处
期刊:Engineering in agriculture, environment and food
[Asian Agricultural and Biological Engineering Association]
日期:2018-07-01
卷期号:11 (3): 127-138
被引量:67
标识
DOI:10.1016/j.eaef.2018.03.001
摘要
Existing agricultural detection algorithms mainly detect a single object category (class) under specific conditions which restricts the farmer's ability to utilize them in different conditions and for different classes. What is needed are generic algorithms that can learn to detect objects from examples, thereby reducing the technical burden required to adapt to local circumstances. Among generic algorithms, deep learning methods recently are beginning to outperform other generic algorithms. In this study, we investigate the possibility of using a deep learning algorithm for recognition of two classes (mature and immature strawberry) of agricultural product using a deep convolutional neural network (DCNN) and greenhouse images taken under natural lighting conditions. To the best of our knowledge, this is the first application of deep learning to the detection of mature and immature strawberries as two classes. We evaluated the results using the following parameters: average precision (AP), a parameter that combines detection success and confidence of detection; and bounding box overlap (BBOL) which measures localization accuracy. The developed deep learning model achieved an AP of 88.03% and 77.21%, and a BBOL of 0.7394 and 0.7045 respectively for mature and immature classes.
科研通智能强力驱动
Strongly Powered by AbleSci AI