1. Preparation of datasets
2. Identification of Mikania micrantha
3. Estimation of invasive plant biomass
We show representative results of a computer vision-based method for the estimation of invasive plants, which is implemented in a programmatic way on a computer. In this experiment, we evaluated the spatial distribution and estimated the biomass of invasive plants in their natural habitats, using Mikania micrantha as a research subject. We utilized a drone camera system to acquire images of the research site, a portion of which is exhibited in Figure 3. We utilized the ResNet101 convolutional neural network to identify the plants present within the study area. Subsequently, we mapped the spatial distribution of invasive plants and illustrated some of our findings in Figure 4. In Figure 3, Mikania micrantha can be observed climbing atop the plant adorned with white flowers. The other plants, as well as the road and accompanying elements, are uniformly depicted in the background. In Figure 4, the model recognizes the red part as Mikania micrantha. Comparing the two sets of images, it is evident that ResNet101 demonstrates robust detection of Mikania micrantha in complex backgrounds. Furthermore, it accurately maps the distribution of Mikania micrantha in the study area with high precision.
The biomass of invasive plants in the study area was estimated by truncating all Mikania micrantha sample plot images from orthophotos at 280 × 280 pixels and extracting the vegetation indices RBRI, GBRI, GRRI, RGRI, NGBDI, and NGBDI. Regression analysis was conducted using the KNNR regression model, with the six indexes as inputs to the estimation model and biomass as the model's output. Figure 5 presents the results: the graph's horizontal coordinates represent the values of the field-measured biomasses, the vertical coordinates represent the values of the model-predicted biomasses, and the gray areas represent the confidence intervals. The results demonstrate strong predictive performance, with an R² value of 0.62 and an RMSE of 10.56 g/m2. The model enhances the accuracy of Mikania micrantha biomass estimation, and the spatial distribution map in Figure 6 effectively captures the distribution of Mikania micrantha biomass.
Figure 1: UAV remote sensing systems. Some examples of RGB image data captured by UAV. Please click here to view a larger version of this figure.
Figure 2: Route planning. Study on regional route planning Please click here to view a larger version of this figure.
Figure 3: Invasive plant identification results within the study area. The figure displays the findings of identifying Mikania micrantha in the study zone via ResNet101 convolutional neural network. The red region in the picture indicates the area detected by ResNet101 as Mikania micrantha, while the other background signifies the rest of the study area. These outcomes correspond to the sample images portrayed in Figure 1. They represent the recognition results of Figure 1, respectively. Please click here to view a larger version of this figure.
Figure 4: Spatial distribution of invasive plants. The model recognizes the red part as Mikania micrantha. Please click here to view a larger version of this figure.
Figure 5: Biomass prediction regression results. The horizontal axis displays biomass values observed in the field, while the vertical axis portrays biomass values estimated by the model. The gray-shaded regions denote confidence intervals. The KNNR model attained an R2 of 0.65 on the test set, while the lowest root mean square error amounted to 30.59 g/m2. In the regression scatter plot of the model, many Mikania micrantha biomass estimations were within the confidence interval, indicating the validity of the biomass prediction. Please click here to view a larger version of this figure.
Figure 6: Spatial distribution of Mikania micrantha biomass. The figure illustrates the estimation of Mikania micrantha biomass throughout the research area utilizing KNNR as the predictive model, along with the extracted Mikania micrantha biomass distribution map. Darker shaded regions represent higher quantities of Mikania micrantha biomass. Please click here to view a larger version of this figure.
Figure 7: Schematic diagram of the main development of this protocol. The figure illustrates the main steps of the protocol presented. Please click here to view a larger version of this figure.
Vegetation Index Name | Calculation Formula |
Green Blue Ratio Index | GBRI = DNG/DNB |
Green Red Ratio Index | GRRI = DNG/DNR |
Red Blue Ratio Index | RBRI = DNR/DNB |
Red Green Ratio Index | RGRI = DNR/DNG |
Normalized Green Blue Difference Index | NGBDI = (DNG – DNB)/(DNG + DNB) |
Normalized Green Red Difference Index | NGRDI = (DNG – DNR)/(DNG + DNR) |
Table 1: Vegetation index calculation formula. The vegetation indices used in this protocol and their respective calculation formulas.
Supplementary Figure 1: Cropping an image to 280 x 280 pixels via Python script using the OpenCV library. Please click here to download this File.
Supplementary Figure 2: Partitioning the dataset into a training set, validation set, and test set. Please click here to download this File.
Supplementary Figure 3: Feature extraction and reduction of dimensionality. (A) Initial feature extraction. (B) Convolution operation. (C) Reduction of dimensionality. Please click here to download this File.
Supplementary Figure 4: Conveying the features to the FC layer in ResNet architecture. Please click here to download this File.
Supplementary Figure 5: Setting the parameters. Please click here to download this File.
Supplemnetary Figure 6: The specific implementation code for generating comprehensive distribution map. Please click here to download this File.
Supplementary Figure 7: RandomResizedCrop and RandomHorizontalFlip functions. Please click here to download this File.
DSLR camera | Nikon | D850 | Sensor type: CMOS; Maximum number of pixels: 46.89 million; Effective number of pixels: 45.75 million; Maximum resolution 8256 x 5504. |
GPU – Graphics Processing Unit | NVIDIA | RTX3090 | |
Hexacopter | DJI | M600PRO | Horizontal flight: 65 km/h (no wind environment); Maximum flight load: 6000 g |
PyCharm | Python IDE | 2023.1 | |
Python | Python | 3.8.0 | |
Pytorch | Pytorch | 1.8.1 |
We report on the detailed steps of a method to estimate the biomass of invasive plants based on UAV remote sensing and computer vision. To collect samples from the study area, we prepared a sample square assembly to randomize the sampling points. An unmanned aerial camera system was constructed using a drone and camera to acquire continuous RGB images of the study area through automated navigation. After completing the shooting, the aboveground biomass in the sample frame was collected, and all correspondences were labeled and packaged. The sample data was processed, and the aerial images were segmented into small images of 280 x 280 pixels to create an image dataset. A deep convolutional neural network was used to map the distribution of Mikania micrantha in the study area, and its vegetation index was obtained. The organisms collected were dried, and the dry weight was recorded as the ground truth biomass. The invasive plant biomass regression model was constructed using the K-nearest neighbor regression (KNNR) by extracting the vegetation index from the sample images as an independent variable and integrating it with the ground truth biomass as a dependent variable. The results showed that it was possible to predict the biomass of invasive plants accurately. An accurate spatial distribution map of invasive plant biomass was generated by image traversal, allowing precise identification of high-risk areas affected by invasive plants. In summary, this study demonstrates the potential of combining unmanned aerial vehicle remote sensing with machine learning techniques to estimate invasive plant biomass. It contributes significantly to the research of new technologies and methods for real-time monitoring of invasive plants and provides technical support for intelligent monitoring and hazard assessment at the regional scale.
We report on the detailed steps of a method to estimate the biomass of invasive plants based on UAV remote sensing and computer vision. To collect samples from the study area, we prepared a sample square assembly to randomize the sampling points. An unmanned aerial camera system was constructed using a drone and camera to acquire continuous RGB images of the study area through automated navigation. After completing the shooting, the aboveground biomass in the sample frame was collected, and all correspondences were labeled and packaged. The sample data was processed, and the aerial images were segmented into small images of 280 x 280 pixels to create an image dataset. A deep convolutional neural network was used to map the distribution of Mikania micrantha in the study area, and its vegetation index was obtained. The organisms collected were dried, and the dry weight was recorded as the ground truth biomass. The invasive plant biomass regression model was constructed using the K-nearest neighbor regression (KNNR) by extracting the vegetation index from the sample images as an independent variable and integrating it with the ground truth biomass as a dependent variable. The results showed that it was possible to predict the biomass of invasive plants accurately. An accurate spatial distribution map of invasive plant biomass was generated by image traversal, allowing precise identification of high-risk areas affected by invasive plants. In summary, this study demonstrates the potential of combining unmanned aerial vehicle remote sensing with machine learning techniques to estimate invasive plant biomass. It contributes significantly to the research of new technologies and methods for real-time monitoring of invasive plants and provides technical support for intelligent monitoring and hazard assessment at the regional scale.
We report on the detailed steps of a method to estimate the biomass of invasive plants based on UAV remote sensing and computer vision. To collect samples from the study area, we prepared a sample square assembly to randomize the sampling points. An unmanned aerial camera system was constructed using a drone and camera to acquire continuous RGB images of the study area through automated navigation. After completing the shooting, the aboveground biomass in the sample frame was collected, and all correspondences were labeled and packaged. The sample data was processed, and the aerial images were segmented into small images of 280 x 280 pixels to create an image dataset. A deep convolutional neural network was used to map the distribution of Mikania micrantha in the study area, and its vegetation index was obtained. The organisms collected were dried, and the dry weight was recorded as the ground truth biomass. The invasive plant biomass regression model was constructed using the K-nearest neighbor regression (KNNR) by extracting the vegetation index from the sample images as an independent variable and integrating it with the ground truth biomass as a dependent variable. The results showed that it was possible to predict the biomass of invasive plants accurately. An accurate spatial distribution map of invasive plant biomass was generated by image traversal, allowing precise identification of high-risk areas affected by invasive plants. In summary, this study demonstrates the potential of combining unmanned aerial vehicle remote sensing with machine learning techniques to estimate invasive plant biomass. It contributes significantly to the research of new technologies and methods for real-time monitoring of invasive plants and provides technical support for intelligent monitoring and hazard assessment at the regional scale.