Precision agriculture involves using technology to optimize farming practices and has given farmers access to more data and greater efficiency for their farms.
With tight profit margins, farmers need to take advantage of technology to lower their costs or increase their revenue.
One area where advancements can prove beneficial is in the measurement of vegetation indices such as the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Red Edge Index (NDRE).
These indices provide information on plant health and can help identify problem areas or where spot applications are needed.
A new paper NDVI/NDRE prediction from standard RGB aerial imagery using deep learning published on Science Direct presents a cost-efficient method for measuring vegetation indices that could help farmers save money, boost efficiency and increase productivity.
Current Methods for Measuring Vegetation Indices
Vegetation indices are reliable indicators for plant health and field assessment, providing important details such as information on growth, vigor, and plant dynamics.
They are designed to maximize sensitivity to vegetation characteristics while minimizing factors such as soil background reflectance and atmospheric effects.
These indices simplify data interpretation and are popular in environmental monitoring, conservation, agriculture, forestry, and related fields.
By using a color map, farmers can view problem spots, varying plant health regions, and other information regarding the field to appropriately assess and manage it.
To measure vegetation indices such as NDVI and NDRE, a multi-thousand dollar multispectral camera is typically used, which is often attached to an unmanned aerial vehicle (UAV) during flight.
The camera captures images of the crops in different spectral bands, and the resulting data is used to calculate the indices.
While these cameras provide accurate data, they come with a significant cost barrier that makes it difficult for most farmers to obtain the technology.
This cost is too expensive, especially for smaller farms or those with tighter budgets.
In addition to the cost, using these cameras can also be time-consuming and require specialized knowledge to operate effectively.
This can make it challenging for farmers to implement this technology, limiting their ability to take advantage of precision agriculture practices.
Using Pix2Pix for Cost-Efficient Prediction of NDVI and NDRE
The study NDVI/NDRE prediction from standard RGB aerial imagery using deep learning published on Science Direct discusses a solution that involves using a conditional Generative Adversarial Network (GAN) called Pix2Pix, along with training data from UAV flights of corn, soybeans, and cotton.
Pix2Pix is a type of machine learning algorithm that can generate images that closely resemble real images, by training Pix2Pix on UAV images and their corresponding NDVI and NDRE values, the algorithm can learn to predict NDVI and NDRE values from RGB images captured by an inexpensive RGB camera.
This low-cost method has the potential to save farmers up to $5000 per UAV system, making precision agriculture more accessible to smaller farmers.
The authors demonstrate that a conditional Generative Adversarial Network (cGAN) called Pix2Pix can create highly comparable NDVI and NDRE color maps.
This is important because remote sensing has seen the rise of machine learning and deep learning, which presents a promising path forward in incorporating artificial intelligence into agriculture and remote sensing.
Generative adversarial networks (GANs) are a type of deep learning that can learn a deep representation without extensively labeled data.
GANs are often used in tasks involving image processing, computer graphics, and computer vision.
They have become increasingly popular in remote sensing, with various applications in image classification, forecasting, and NDVI prediction.
The unique aspect of the research presented in this paper is the use of Pix2Pix with aerial imagery collected from small UAV flights.
Pix2Pix is a conditional GAN designed to produce artificial images that are indistinguishable from real images which are achieved by feeding the data that is desired to be conditioned onto the generator and the discriminator.
With the ability to accurately predict NDVI and NDRE values using a low-cost RGB camera, farmers can identify problem areas and optimize their farming practices, leading to greater efficiency and potentially higher profits.