| 摘要: |
| 针对传统方法对苹果叶片进行图像分割和测量几何形状参数精确度较低的问题，结合基于深度学习和引导滤波技术提出一种新的苹果叶片图像自动分割算法。首先采用深度学习方法，使用BiseNet卷积神经网络对苹果叶片图像进行自动分割，得到苹果叶片主体轮廓；然后使用彩色苹果叶片图像作为引导图像对主体轮廓进行引导滤波处理，以增强边缘锯齿等细节特征信息；最后将主体轮廓与细节特征信息进行联合分割，得到完整、准确的苹果叶片信息。对包含174种8 184张苹果叶片图像数据集进行试验，结果表明苹果叶片分割的精确率达到98.99%，交并比98.82%。利用本研究算法能够真正实现准确、快速测量苹果叶片的面积、周长等参数值，为苹果叶片几何参数的测定提供了一种新的测量方法。 |
| 关键词: 苹果叶片 图像分割 深度学习 引导滤波 |
| DOI：10.11841/j.issn.1007-4333.2020.09.12 |
| 投稿时间：2020-02-05 |
| 基金项目:国家自然科学基金项目(41773024,U1404103)；河南省高等学校重点科研项目(18B520017) |
|
| An apple leaf image segmentation algorithm using deep learning and guided filtering |
|
ZHENG Yanmei,HUANG Guangyao,LU Bibo,WANG Yongmao
|
| (College of Computer Science and Technology, Henan Polytechnic University, Jiaozuo 454003, China) |
| Abstract: |
| Aiming at the low accuracy of image segmentation and geometric parameter measurement of apple leaves by traditional methods, a new automatic image segmentation algorithm of apple leaves based on deep learning and guided filtering technology is proposed. Deep learning method is firstly adopted in this study. BiseNet convolutional neural network is used to automatically segment the image of apple leaves to obtain the main contour of apple leaves. The color apple leaf image is then used as the guide image to conduct the guiding filtering processing on the main body contour to enhance the detail feature information such as sawtooth edge. Finally, the outline of the main body and the detailed feature information were combined to segment to obtain complete and accurate information of apple leaves. The results of 8184 apple leaf image data sets containing 174 kinds of apple leaves show that the accuracy of apple leaf segmentation and IoU ratio of the proposed algorithm are 98. 99% and 98. 82% respectively. In conclusion, the area and circumference of apple leaves can be measured accurately and quickly by using the new algorithm established in this study. This study provides a new method for the geometric parameter measurements of apple leaves |
| Key words: apple leaves image segmentation deep learning guided filtering |
