International Journal of Circuits, Systems and Signal Processing

E-ISSN: 1998-4464
Volume 15, 2021

Notice: As of 2014 and for the forthcoming years, the publication frequency/periodicity of NAUN Journals is adapted to the 'continuously updated' model. What this means is that instead of being separated into issues, new papers will be added on a continuous basis, allowing a more regular flow and shorter publication times. The papers will appear in reverse order, therefore the most recent one will be on top.

Main Page

Submit a paper | Submission terms | Paper format


Volume 15, 2021

Title of the Paper: Spruce Counting Based on Lightweight Mask R-CNN with UAV Images


Authors: Wenjing Zhou, Xueyan Zhu, Mengmeng Gu, Fengjun Chen

Pages: 634-642 

DOI: 10.46300/9106.2021.15.70     XML


Abstract: To achieve rapid and accurate counting of seedlings on mobile terminals such as Unmanned Aerial Vehicle (UAV), we propose a lightweight spruce counting model. Given the difficulties of spruce adhesion and complex environment interference, we adopt the Mask R-CNN as the basic model, which performs instance-level segmentation of the target. To successfully apply the basic model to the mobile terminal applications, we modify the Mask R-CNN model in terms of the light-weighted as follows: the feature extraction network is changed to MobileNetV1 network; NMS is changed to Fast NMS. At the implementation level, we expand the 403 spruce images taken by UAV to the 1612 images, where 1440 images are selected as the training set and 172 images are selected as the test set. We evaluate the lightweight Mask R-CNN model. Experimental results indicate that the Mean Counting Accuracy (MCA) is 95%, the Mean Absolute Error (MAE) is 8.02, the Mean Square Error (MSE) is 181.55, the Average Counting Time (ACT) is 1.514 s, and the Model Size (MS) is 90Mb. We compare the lightweight Mask R-CNN model with the counting effects of the Mask R-CNN model, the SSD+MobileNetV1 counting model, the FCN+Hough circle counting model, and the FCN+Slice counting model. ACT of the lightweight Mask R-CNN model is 0.876 s, 0.359 s, 1.691 s, and 2.443 s faster than the other four models, respectively. In terms of MCA, the lightweight Mask R-CNN model is similar to the Mask R-CNN model. It is 4.2%, 5.2%, and 9.3% higher than the SSD+MobileNetV1 counting model, the FCN+Slice counting model, and the FCN+Hough circle counting model, respectively. Experimental results demonstrate that the lightweight Mask R-CNN model achieves high accuracy and real-time performance, and makes a valuable exploration for the deployment of automatic seedling counting on the mobile terminal.