1 材料与方法
1.1 图像采集
1.2 方法
2 结果与分析
图6 基于Faster R-CNN模型的芒果图像识别效果A:低挂果密度整株芒果树图;B:低挂果密度整株芒果树识别效果图;C:高挂果密度整株芒果树图;D:高挂果密度整株芒果树识别效果图;E:芒果植株树冠挂果图;F:芒果植株树冠挂果识别效果图。 Fig. 6 Mango image detection results by Faster R-CNN A: The image of a whole mango tree with low fructification density; B: The image detection results of a whole mango tree with low fructification density; C: The image of a whole mango tree with high fructification density; D: The image detection results of a whole mango tree with high fructification density; E: The crown image of a mango tree; F: The image detection results of a mango tree crown. |
表1 芒果识别计数准确率分析Tab. 1 Mango image recognition and counting performance by Faster R-CNN |
| 统计量Statistics | 值Value | 统计量Statistics | 值Value |
|---|---|---|---|
| 人工计数 | 2291 | 平均计数误差 | 4.2 |
| 检出个数 | 1968 | 平均计数误差率 | 7.9% |
| 正确识别并计数个数 | 1619 | 计数识别准确率 | 82.3% |
| 漏检个数 | 267 | 漏检率 | 11.7% |
| 误检个数 | 198 | 误检率 | 8.6% |
