Using Household Drones to Determine Agricultural Health
Introduction
Agriculture is incredibly important to society, but growing crops can often prove incredibly difficult, particularly for small farms that don’t have the resources to ensure they are growing their crops on well-suited land. While there are some ways to solve this problem for large farms that can afford it, there is no good universal solution to evaluating soil and ensuring your farm is land that can support crops.
We were inspired by our collective interest in drones and the role they were capable of playing in providing opportunities to smaller-scale farmers. Our app appeals to those without the financial means to hire professional surveyors or buy specialized agricultural drones, as all of our services work with standard off-the-shelf drones. We also don't require our customers to be drone-savvy whatsoever -- we devised a path-planning app capable of connecting to a drone's SDK and flying it for the user.
The vision for our app included two main components: a web analyzer and a mobile path planner. Users would be able to select a region that they wanted to map out using the path planner. The app would then auto-generate a path for the drone to fly that fully covered the necessary area, while also displaying information like the current weather and the estimated time/battery loss from the flight. The user could then press a button to launch the drone, and the path data would be used to automatically fly the drone using the DJI SDK.
The web analyzer would then take a synthesized image from the result of the drone flight and use AI to determine regions of high and low soil quality, trained on datasets where the knowledge of quality and nutrient concentration for a given region is known. Notable regions of the image would be highlighted and accompanied by pop-ups, and a sidebar would provide a summary of either the entire image or the selected region. That summary would include information like soil health, saturation, and individual nutrient concentration.
Our Process
The four of us kicked off our brainstorming process with Figma, which we used to devise initial designs for our website and mobile app that we closely matched with our actual interfaces once we transitioned into development. Once we were satisfied with our design, we broke off into our separate tasks while regularly conferring to hold each other accountable and help each other when needed. We all shared a single git repository with separate folders for our web and mobile elements that we regularly pushed and pulled from to work collaboratively.
Our path planner uses Flutter and pulls from the Google Maps API. Users are able to view their current GPS location and drag two control points to change the size and location of the region they want to film. In the future, we hope complete path auto-generation and successfully interface with the drone SDK to automatically fly it.
We built our web analyzer using Svelte.js. It provides useful information describing areas of interest, presenting detailed information about soil quality and nutrient availability for a given image. Our next step for this component is completing the AI model that will analyze any image and update our data accordingly.
Accessibility
How do we ensure everyone has access to our solution?
Providing mass accessibility to professional tools can be incredibly difficult because the technology involved is often very advanced and expensive. It also usually requires companies to go out to take soil samples of various locations and verify if the contents are suitable for crop growth. This often isn’t accessible for many farmers. We aim to solve this problem by using DJI Drones. By using off-the-shelf drones, we can extract this extremely important information, and allow farmers to monitor the quality of their farmland.
The DJI Mini SE is a $300 Drone that supports our software solution, and it doesn’t have to be registered with the FAA due to it being 249 grams, making it an ideal target for accessibility. These drones can be controlled by the DJI SDK, which allows the user to input coordinates that the drone can follow. We can combine this in conjunction with an easy-to-use interface to select a box around an area you want to be surveyed, and we can automatically generate an optimal path to capture the terrain.
Data Processing
Importing the Drone Data
After capturing data from the drone, we’ll need to have a way to parse that data and produce an actual analysis. We do this in a multi-stage process- first, we need to identify the land mass. This can be done using Gaussian Blur and OpenCV for edge detection. This allows us to extract the important parts of the terrain where farming is a consideration.
Elevation / Runoff
Checking topography is extremely important in determining whether the land is suitable for planting crops in the first place. If there is a major concern for runoff into a particular region of land, that should be avoided to ensure crops don’t get washed away, oversaturated, covered in above sediment, or receive too much nutrients. We first take the logs from the drone, find the elevation of every chunk of data in the dataset, and parse this data to find possible runoff areas, where there are major concerns of soil oversaturated with fertilizer and nutrients from above.
Soil Health
Before planting crops in any region, we want to ensure the soil has the appropriate contents for whatever crop you are trying to grow. Because the needs of each individual crop vary, we need to use optical recognition paired with Artificial Intelligence and Machine Learning to estimate the contents of the soil, as well as the overall health. According to ScienceDirect, “Soil color is influenced by its mineral composition as well as water and organic contents. For example, soils high in calcium tend to be white, those high in iron reddish, and those high in humus dark brown to black. Soil needs only about 5% organic material to appear black when wet.” We can use this principle to determine the contents of the soil without having to physically extract and examine samples from various parts of the region.
Figure 1: “Soil color and spectroradiometry”, ScienceDirect.com
According to Figure 1, we can use The Munsell System of Color Notation to quantitatively measure soil color. We can use the optical properties of soil to categorize it in such a way as to calculate its contents.
Figure 2. “Spectral reflectance curves for Newtonia silt loam at various moisture contents,” ScienceDirect.com
We can use the principles behind Figure 2 to determine various properties of soil, which can then be used to analyze various properties of soil.
Crop Health
We can use a similar process to monitor the health of crops after they are grown. By using OpenCV / Machine Learning, we can keep track of the health of individual crops automatically. This can help farmers identify that there is a problem with their crops, and in some cases even identify the exact cause of it.
Visualization
Landscape Filtering
A major difficulty in capturing large landscapes and parsing them is filtering out unimportant edges. There are many cases where a body of water may be present on land, which is something we want to avoid parsing to avoid poisoning the actual conclusion.
Figure 3. Filtering out a large body of water.
We can use a mix of numpy, OpenCV, and matplotlib to parse this data out. For this, I used “Canny”, a portion of OpenCV that allows you to take optical data as input and filters out the terrain. This is an imperfect example, but it contains enough accurate data that we can use it after minimal post-processing. This was done with a simple function in Python:
import cv2
import numpy as np
from matplotlib import pyplot as plt
image = cv2.imread('coastline.png')
edges = cv2.Canny(image,100,200)
plt.subplot(121),plt.imshow(image,cmap = 'gray')
plt.title('Original Landscape')
plt.xticks([])
plt.yticks([])
plt.subplot(122),plt.imshow(edges,cmap = 'gray')
plt.title('Filtered Landscape')
plt.xticks([])
plt.yticks([])
plt.show()
Sources
https://www.sciencedirect.com/topics/agricultural-and-biological-sciences/soil-color
https://www.opencvhelp.org/tutorials/applications/agriculture-and-farming/
Log in or sign up for Devpost to join the conversation.