INVASIVE SPECIES MAPPING USING LOW COST HYPERSPECTRAL IMAGERY

Steven Jay ¹ â€“ Research Assistant

Dr. Rick Lawrence ¹ â€“ Associate Professor

Dr. Kevin Repasky ² â€“ Associate Professor

Charlie Keith ² â€“ Research Assistant

¹ Department of Land Resources and Environmental Science Montana State University â€“ Bozeman

² Department of Electrical & Computer Engineering Montana State University â€“ Bozeman 128 AJM Hall Montana State University Bozeman, MT 59717

Monitoring of invasive species has long been a time consuming, expensive and ineffective job. Remote sensing is a means of monitoring invasive species, however, it is limited by funding, time and accuracy issues.

This study evaluated a cost-effective hyperspectral imager that monitors and distinguishes Euphorbia esula located in the grassland ecosystem. Ground images were collected weekly from the summer of 2008 to identify phenological periods in which the milk scorpion can be accurately monitored and imaged. Weekly images acquired by hyperspectral imagers can also be used to evaluate time series classifications, while traditional methods require significant expense. In this study, a random forest model was used for image classification. The classification accuracy was compared with the data obtained by traditional laboratory analysis methods. Multi-time classification can improve classification accuracy. Future research will be combined with the device for low-cost and efficient invasive species monitoring on drones or low-altitude aircraft.

Subjects:

Latex sputum: Eurasia invades plants and is toxic. It grows in bare earth, floods in pastures, wasteland, fallow farmland, and urgently need management measures to reduce grazing and farmland losses. Traditional herbicide methods are ineffective, and sheep or artificial weeding is more effective.

Problems in traditional method monitoring:

The use of ground observation methods is difficult to monitor and expensive, so it is difficult to evaluate the effectiveness of management measures. The combination of GPS and GIS is effective for monitoring and mapping, but it requires a lot of funding and technical support.

Remote sensing has become a widely used method for studying invasive plants. Since images can cover larger research areas, block data not detected by ground surveys can be obtained. However, the relative cost of drawing is relatively high, and some satellite sensors have low spatial resolution and spectral resolution, limited time resolution, and problematic accuracy, and data analysis and integration require a strong technical team.

Remote sensing methods to monitor invasive species:

High spectral resolution, low spatial resolution - hyperspectral sensor

Collecting a large amount of spectral information, but the price is expensive, data processing is difficult, and the application is limited.

High spatial resolution, low spectral resolution

Random forest classification models are used to classify multispectral and hyperspectral images. Suitable for uniform plant communities, as well as for mixing plant communities.

Multi-period hyperspectral image data can reflect the phenological period of the plant.

Research methods:

Instrument selection:

Pika II (Resonon, USA)

400-900nm, resolution: 2.1nm, bit depth: 12, frame rate: 60/s

Get an image every 7-10 days (both at 9-11 am) and get 11 images

Reference cloth: 2x2m, 3 blue reference cloth, pixel resolution 5cm

How to measure?

PIKA II: Fixed on a mountain top with a tripod, artificially timed to capture hyperspectral images of the valley area

Ground measurement: randomly set 35 reference points, small flag markers, GPS information,

1m sample survey, vegetation density, 0-174 plants/m ²

Cell resolution: calculated by measuring the number of pixels between the reference point in the image and the measured GPS position;

Resolution check: check with the pixels obtained by the 2x2m reference cloth;

Raw data: extract and average the reference cloth to a pixel within 0.5m radius as practice data;

Using the random forest algorithm to analyze each image, processing each pixel, distinguishing between the presence of large ticks and the absence of large ticks, the pixels are matched with the original position, and the image regions with large ticks and no large ticks are drawn. ;

Single-time image classification: It is effective to use this technology to monitor the grassland. The estimated accuracy of 11 different time acquisitions is 72%-95%. The best estimated time is in early July. The flowering period reaches its peak, and it begins to age in the later stage, which is difficult to identify in the early stage.

in conclusion:

1. The preliminary conclusions of this study are very promising, as the data for a single day can be matched to other hyperspectral measurements, and further research on existing analyses and reduction of errors will make the results more valuable.

2. The development of this cost-effective hyperspectral sensor will be a breakthrough in hyperspectral remote sensing science and will open up new horizons for the application of hyperspectral data.

3. Accurate, robust, and low-cost hyperspectral sensors will provide a broader opportunity for potentially multi-time scale hyperspectral analysis and multi-platform applications such as drones and light aircraft.

Click here to view the original literature

Hand Sanitizer

75% Alcohol Gentle Hand Sanitizer Hand Disinfection Gel

Hand Sanitizer

Luck Medical Consumables Co.,LIMITED , https://www.luckmedical.com