ANDS Logo

Data collections can be seen on:

https://researchdata.ands.org.au/mango-fruit-tree-image-collection/815790

Data Management Policy/Procedure:

https://my.cqu.edu.au/web/eresearch/research-data-management

Project Members:

Libbie Blanchard (Project Manager, e.blanchard@cqu.edu.au)

Prof Kerry B. Walsh (Lead Researcher, k.walsh@cqu.edu.au)

ANDS Contact:

Kathryn Unsworth (kathryn.unsworth@ands.org.au)

Project Status:

Completed

Central Queensland University: Mango fruit on tree image collection

Central Queensland University

Project Description

The Institute for Future Farming Systems research focuses on tropical and subtropical agricultural systems, and operates in a region where collaborators can work with industry partners in best practice agricultural production systems.

This project aims to develop a machine vision based estimate of mango crop load up to 4 weeks before harvest, to inform farm manager decisions on labour hire, packhouse consumable purchasing etc.

Crop yield estimation in mango orchards is currently undertaken on best practice farms only as it is a manual process involving many man hours. A human operator uses a hand counter to estimate fruit number on up to 2% of trees in the orchard. The use of a machine vision based system could replace this manual system and improve predictions for harvesting and marketing requirements.
Machine vision systems are developed using a training set of images. Unfortunately, unlike a factory process line, orchards are variable, in tree shape, leaf colour, fruit size, fruit colour, etc., and lighting conditions can also be variable. Thus it is important to test a machine vision system, once trained, across a number of new image sets. To compare machine vision algorithms, it is very useful to use the same populations of training and validation images.

This collection of images of mango trees with fruit at stone hardening stage under artificial illumination has been collected from different areas on one farm, from different seasons and from different farms/growing areas. It has been used to compare different machine vision algorithms (e,g, Waqar et al., 2014), and is made available for other researchers to trial their machine vision approaches, benchmarking their results against that of Waqar et al. (2014).

Further image sets will be added in the future. In particular it is acknowledged that machine learning techniques require use of many (thousands) of labelled vignetted images.

Data Type:

Image files - 3 sets: - jpg - jpg and count filter from Photoshop - raw CR2

ANZSRC-FOR code:

070601
070101