Fig. 1. The acquisition system based on a multispectral camera
(a): The composition of the acquisition system; (b): DJI Phantom; (c): ADC Micro
1: DJI Phantom, a quad-rotor UAV; 2: ADC Micro, a multispectral camera; 3: an SD card with 16G storage; 4: Computer
Fig. 2. The acquisition of multispectral images in fields
(a): Experimental site; (b): Sampling scheme
Fig. 3. The acquired multispectral images
(a): An image example (1); (b): An image example (2);(c): An image example (3); (d): An image example (4)
Fig. 4. The examples of the images with R, G and NIR components processed by PixelWrench2 x64
(a): The image with R component;(b): The image with G component;(c): The image with Near-Infrared (NIR) conponent
Fig. 5. The results of R, G and NIR components
(a): R, G and NIR components in local areas;(b): R, G and NIR components in entire areas
Fig. 6. The portable sugar meter, PAL-1, and its real measurement
(a): A portable sugar meter, PAL-1;(b): Measuring total sugar of grape juice by using PAL-1
Fig. 7. The changing relation between each component and date
(a): The changing relation between R component and date; (b): The changing relation between G component and date;(c): The changing relation between NIR component and date
Fig. 8. The regression modelling between the local R component of the multispectral images and total sugar
(a): The linear model between the local R component of the multispectral images and total sugar;(b): The logarithmic model between the local R component of the multispectral images and total sugar
日期 | 局部R 分量 | 局部G 分量 | 局部NIR 分量 | 整体R 分量 | 整体G 分量 | 整体NIR 分量 | 检测 序数 |
---|
2016.9.24 | 6.5 | 52.3 | 163.8 | 7.0 | 44.5 | 136.7 | 1 | 2016.9.26 | 3.8 | 48 | 140.0 | 5.4 | 40.8 | 113.8 | 2 | 2016.9.28 | 8.8 | 47.8 | 156.7 | 12.2 | 39.8 | 123.0 | 3 | 2016.9.30 | 8.3 | 50.8 | 149.5 | 10.0 | 41.2 | 117.3 | 4 | 2016.10.3 | 10.2 | 53.8 | 161.7 | 8.3 | 40.0 | 123.3 | 5 | 2016.10.5 | 14.7 | 47.4 | 148.9 | 12.1 | 36.0 | 113.3 | 6 | 2016.10.8 | 16.5 | 61.7 | 144.3 | 16.0 | 49.7 | 110.8 | 7 | 2016.10.10 | 29.0 | 63.3 | 135.3 | 28.0 | 45.5 | 91.0 | 8 | 2016.10.12 | 32.6 | 81.6 | 128.4 | 27.6 | 55.6 | 128.4 | 9 |
|
Table 1. The component results of local and entire areas
日期 | 总糖含量/% | 检测序数 |
---|
2016.9.24 | 18.4 | 1 | 2016.9.26 | 17.1 | 2 | 2016.9.28 | 19.9 | 3 | 2016.9.30 | 19.7 | 4 | 2016.10.3 | 19.9 | 5 | 2016.10.5 | 20.8 | 6 | 2016.10.8 | 21.6 | 7 | 2016.10.10 | 22.0 | 8 | 2016.10.12 | 21.9 | 9 |
|
Table 2. The total sugar of grape juice of the model set
颜色分量 | 回归方程 | 调整后R2值 | F值 | p-value | 显著性 |
---|
局部R分量 | y=3.363x-2.338 | 0.815 | 36.2431 1 | 5.314 44×10-4 | ***** | 整体R分量 | y=2.665x+0.742 | 0.721 83 | 21.759 51 | 0.002 3 | *** | 局部G分量 | y=3.125x+40.675 | 0.532 8 | 10.123 25 | 0.015 45 | ** | 整体G分量 | y=1.218x+37.586 | 0.215 | 3.192 52 | 0.117 13 | - | 局部NIR | y=-3.019x+162.714 | 0.406 13 | 6.470 9 | 0.038 45 | * | 整体NIR | y=-2.167x+128.344 | 0.099 2 | 1.881 03 | 0.212 56 | - |
|
Table 3. The regression of each colour component
拟合方式 | 拟合公式 | 调整后R2 | F值 | p-value | 显著性水平 | 备注 |
---|
线性拟合 | y=0.140 1x+18.115 | 0.694 19 | 19.159 76 | 0.003 25 | *** | 80%数据集 | 对数拟合 | y=22.194- | 0.970 62 | 11 522.402 65 | 5.124 07×10-10 | ***** | 80%数据集 |
|
Table 4. The results of the regression analysis between the local R component of the multispectral images and total sugar
日期 | 多光谱 图像R 分量 | 模型预测 总糖含量 /% | 实际检测 总糖含量 /% | 误差 /% |
---|
2016年9月24日 | 5.7 | 18.2 | 18.4 | -1.09 | 2016年9月26日 | 4.0 | 17.2 | 17.1 | +0.58 | 2016年9月28日 | 9.7 | 20.0 | 19.9 | +0.50 | 2016年9月30日 | 11.7 | 20.6 | 19.7 | +4.57 | 2016年10月3日 | 8.3 | 19.5 | 19.9 | -2.01 | 2016年10月5日 | 11.3 | 20.5 | 20.8 | -1.44 | 2016年10月8日 | 16.0 | 21.3 | 21.6 | -1.39 | 2016年10月10日 | 31.2 | 21.9 | 22.0 | -0.45 | 2016年10月12日 | 31.5 | 22.0 | 21.9 | +0.46 |
|
Table 5. The comparison between the total sugar of the validation set and that of the model prediction results