Collecting and Processing Data on a 3000Ha Cocoa Field for Automated Crop Monitoring

Overview

Client X is a large agricultural company that wished to optimize its practices in crop monitoring, early disease detection, and weed identification by implementing AgTech solutions. The pilot project was to be conducted on a cocoa plantation covering a 3000 ha area. The Client did not have an in-house team of data scientists to acquire or annotate the data to be used as training datasets.

They approached Aya Data to come up with a cost-effective and practical solution.

Services

Image segmentation

Industry

Manufacturing

Headquarters

London, UK

Company Size

250+

The Challenge

  1. Traditional methods of data collection for large crop fields and the accompanying analyses are time-consuming, labor-intensive, and lack comprehensive coverage.
  2. Image and video annotation techniques that are used to annotate large volumes of data are imprecise, often leading to subpar computer vision models.
Samuel Sundin
CCO
About
Chief Commercial Officer (CCO) - Sam has a wealth of experience across the technology and AI value chain, with a career forged at Microsoft, IBM and Cloudfactory amongst others. As CCO at Aya Sam has a simple remit to build long term, mutually beneficial relationships with businesses looking to access the power of AI.

The Solution

Drones equipped with RGB cameras, LiDAR sensors, and multispectral sensors were to be programmed to autonomously capture high-resolution RGB imagery, LiDAR data, and multispectral data across the 3000 ha cocoa plantation area.

The collected data would be processed, with human oversight, into tiles by Aya Data, with the main criteria for individual tiles being geospatial coordinates

The Results

RGB Imagery: The high-resolution imagery provided detailed visual information about the cocoa plants, aiding in plantation mapping, plant health assessment, and identification of potential issues.

LiDAR Data: The LiDAR data facilitated precise modeling of the cocoa plantation’s topography, tree heights, and structural attributes. It enabled accurate measurement of tree heights and identification of canopy gaps or uneven growth patterns.

Multispectral Data: The multispectral data allowed for the assessment of cocoa plant health by analyzing various spectral bands, such as NIR and TIR. Vegetation indices like NDVI and NDWI were derived to monitor crop health, identify stress areas, and optimize resource allocation.

The collected data was then processed into tiles by Aya Data’s full-time annotation staff, providing the client with structured data which could further be processed into training datasets for CV models.

At completing the pilot project, the Client proposed that Aya Data create a bespoke computer vision model that could automatically identify diseases in cocoa plants and flag areas with high weed density that could stymie the growth of the plant each time drones were set to capture video and images of a given cocoa plantation.

Building and deploying machine learning models that could automatically monitor additional crop health and stress parameters are also in discussion.