Treffer: 单子叶作物叶片气孔自动识别与计数技术.

Stomata is the dominant gate for water and gas exchange for plant leaves, and thus plays key roles in plants in response to the fluctuations of the environmental variables. Observation and counting stomata amounts is generally one of the obligatory determinants in the research of plant ecology and physiology. The classic measurement protocol of leaf stomata usually includes the capture of leaf stomata by a microscope, followed by manually identifying and counting of the target stomata. This method is well-known in disadvantages of both time and labor consuming, and of low accuracy. Although some algorithms for stomata recognition have been proposed at present, their recognition abilities showed limitation, and they could not realize the full effect of automatic recognition. Thereafter, we developed an automatic identification and counting technique based on YOLOv3, one of the high speed convolutional neural networks (CNN) algorithm in the present study. We acquired pictures of leaf stomata after the third leaf occurred and during grain filling stage of wheat (Triticum aestivum), in which, 138 pictures were taken from the method of nail polish printing, and another 117 pictures were taken from the method of portable microscopy. After that, we created separate data sets and then trained the corresponding models respectively. During the training process, we visualized the loss and average loss which were the most important training parameters, and finally stopped the training at 1 200 times. To better describe the parameters of both models, we used the key metrics to evaluate the models, such as precision, recall and F1. The precision, recall and F1 reached 0.96, 0.98 and 0.97 in the method of nail polish printing, whereas reached 0.95, 0.98 and 0.96 in the method of portable microscopy. Secondly, this algorithm could count stomata amounts accurately, and showed excellent robustness. By linear regression between the labeled and predicted stomata amounts in pictures from test sets, we found that this algorithm showed strong correlation, R² were 0.980 1 and 0.962 5, respectively. What’s more, this algorithm also showed high performance in high-throughput and real-time, since it identified stomata with a speed of 30 frames per second. With this technique, we optimized the objective identifying performance, which conferred accurate identification performance of stomata in the microscope pictures of leaf stomata. Firstly, compared with the method of nail polish printing, the method of portable microscopy showed low precision and F1, but was harmless to samples. Secondly, YOLOv3 algorithm exhibited the merits of accuracy, high efficiency, as well as real-time, long-time and dynamic detection. Thirdly, this technique was high compatible due to its power in accurately identifying stomata of other monocotyledonous crops such as barley (Hordeum vulgare), rice (Oryza sativa) and maize (Zea mays). Lastly, in order to facilitate the use of more researchers, we not only opened source of the detailed Python code, but also encapsulated the method in a relatively complete way. It could provide an interface for relevant researchers to detect their stomata photo or video files. The files in our stomata project could be consulted and downloaded in Github (https://github.com/shem123456/). [ABSTRACT FROM AUTHOR]

为实现作物叶片气孔的自动识别与快速计数，该研究采用卷积神经网络中高计算效率的YOLOv3 算法，开发了 一种全自动气孔识别和计数解决方案。该算法优化了物体检测性能，可准确识别显微图像中的气孔。其中，对指甲油印 迹法获得照片的气孔检测精确率、召回率和F1 值分别为0.96，0.98 和0.97，便携式显微镜拍摄法照片气孔检测精确率、 召回率和F1 值分别为0.95，0.98 和0.96，具有很好的鲁棒性。该算法检测速度快，可实现对30 帧/s 的视频文件进行快 速气孔识别，实现了实时检测。此外，采用拍摄的小麦叶片照片进行训练得到的气孔识别模型，还可同时实现对大麦、 水稻和玉米等单子叶作物叶片气孔的识别，其中，大麦的检测精确率、召回率和F1 值分别为0.94，0.83 和0.88；水稻的 检测精确率、召回率和F1 值分别为0.89，0.42 和0.57；玉米的检测精确率、召回率和F1 值分别为0.91、0.76 和0.83； 显示出模型良好的泛化能力。 [ABSTRACT FROM AUTHOR]

Copyright of Transactions of the Chinese Society of Agricultural Engineering is the property of Chinese Society of Agricultural Engineering and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)