Developing Machine Learning Methods for RGB Images to Quantify Crop and Weed Populations Across Agricultural Fields

Food security is a global concern due to the ever-increasing population amid limited agricultural land resources. Farmers utilize fertilizers and herbicides to ensure consistent crop establishment. However, overuse of fertilizers and other treatments leads to negative consequences resulting in soil depletion, environmental contamination, and pollution. An effective way to reduce these toxic chemicals is to apply the optimal amount based on crop/weed distribution. Uneven crop distribution adversely impacts crop production, and it is important to identify inconsistent areas within a field. Once inconsistent regions are identified, site-specific treatments can be prescribed. Site-
specific farm management practices require high-resolution quantification of crop establishment like plant stand counting, leaf area estimation, inter-row spacing, inter-plant spacing, plants per crop row and plants per acre. Traditionally, manual scouting is performed to estimate these parameters, which is subjective, time-consuming, costly and not fully representative of the whole field. In this research proposal, we develop methods for quantifying crop establishment and biotic stress mapping using AI methods. Recently, highly accurate deep learning-based object detection methods have been developed. These methods are widely applied for non-agriculture applications like autonomous vehicles and medical imaging.

Faculty Supervisor:

Abdul Bais

Student:

Partner:

Croptimistic Technology Inc

Discipline:

Engineering

Sector:

Agriculture; Professional, scientific and technical services

University:

University of Regina

Program: