Fruit Crop Prediction And Biomass Calculation

About the Client

A company in Central America that provides scientific activities in AgriTech to revolutionize the way to monitor and control crop fields and plantations.

Business Challenge

Accurate crop yield predicting ahead of its harvest time has a significant implication for agricultural businesses letting them measure and forecast selling and storage capacities as well as be efficient with crop production measurement. Our client wanted to have a desktop application that could accept drones images as input and represent results in GUI with an ability to select a specific area. Using the analyzed data, the client wanted to plan his resources and budget more efficiently, which would allow him to save costs and implement this knowledge into managing other fields.

Solution Overview

The client has an Agritech platform that contains data collected from agricultural drones. The desktop application developed by Quantum should help the client to calculate biomass in kilos and obtain a specific number of fruits from an image per field (or any specified area, selected on the field) and further analyze the available data by means of computer vision, and artificial intelligence to predict fruit ripeness, which will allow efficiently plan next steps as to biomass managing.

Project Description

The project consisted of several parts that included data collection, pre-processing and analysis. As mentioned above, the source of data was agricultural drones possessed by the client. The available data was collected and cleansed, which allowed us to start developing an automatic fruit recognition system.  Based on DEM (digital elevation model) data we have built a model that identifies a plant’s height. Using LAS data we have extracted points from data and merged them into objects. We have successfully classified biomass and divided it into two parts – usable (fruit and hay) and futile (to be disposed). Based on this, we could predict the amount of biomass that grows on the field under the same conditions.

One of the challenging tasks was to identify and accurately calculate plant height. Drones imagery had a lot of artifacts, such as shadows or low images overlap. As a result, it was difficult to distinguish a small plant from its bigger neighbour, as it was covered by the shadow of the bigger plant. Increasing  the number of flights over the field under different daylight conditions has helped to overcome this issue and deliver feasible results that helped the client to meticulously plan his further steps as to biomass managing.

Let's discuss your idea!

pineapples_1

Technological Details

To calculate the plans’ height, process imagery and extract data from it we used the Open Source Geospatial Foundation (OSGEO) tools. The following technology stack was used: GDAL, numpy, shapely, geopandas, laspy, etc. We have used  QGIS for research, visualizations and verifications. We also used pix4d as a part of processing pipeline.

gdal
gdal
Numpy
Numpy
geopandas
geopandas
OSGEO
OSGEO
Shapely
Shapely
QGIS
QGIS
pix4d
pix4d

Connect with our experts