McMahon, Sean (2018) Camera-based plant sensing system for estimating crop maturity. [USQ Project]
Abstract
The vegetable industry is a significant part of Australia’s agricultural sector. Environmental and economic benefits can occur within the industry through having tools to help improve efficiencies with irrigation, fertiliser and harvest management, as well as make better decisions relating to markets / fulfilling contracts. The most common methods of crop maturity estimation are through field observations and crop sampling along with experience of the grower. Research has been carried on various vegetable crops to estimate crop maturity, although limited research has been carried out on root vegetables who’s harvestable section grows below ground level. Therefore, this project aimed to develop a camera-based plant sensing system for estimating carrot crop maturity in a variety of soil types and management practices.
Imagery and field data were collected throughout the growing season in a carrot crop at Kalbar, Fassifern Valley. A colour thresholding image analysis algorithm was developed for canopy size detection. This algorithm was created utilising the ’open cv’ computer vision package within the Python programming language. Machine learning regression and classification models were developed for estimating crop maturity. Various regression models were compared and tuned using Weka data mining software. A K-Nearest Neighbours regression algorithm was found to be the best performing regression model. While the regression model is a more suitable for maturity estimation, the overall performance of the model was not suitable as a final model. Various machine learning classifier models were then compared using the ’sci-kit learn’ machine learning library within the Python programming language. A K-Nearest Neighbours classifier was found to be the best performing classifier model. The results were found to be reasonably accurate although the classifications were quite broad and not as useful as the regression model. The major limitation to both models was the small size of the data-set. This resulted in poor performance of the regression model and probable overfitting of the classifier model.
Further research involving larger data-sets over multiple seasons and at multiple locations would improve the outcomes of both types of models and allow the model to be more transferable to different soil types and climates. In addition, further development and optimisation of the code is required to develop a system for on-the-go data collection.
|
Statistics for this ePrint Item |
| Item Type: | USQ Project |
|---|---|
| Item Status: | Live Archive |
| Faculty/School / Institute/Centre: | Historic - Faculty of Health, Engineering and Sciences - School of Civil Engineering and Surveying (1 Jul 2013 - 31 Dec 2021) |
| Supervisors: | McCarthy, Alison |
| Qualification: | Bachelor of Engineering (Agricultural) |
| Date Deposited: | 02 Sep 2022 00:12 |
| Last Modified: | 29 Jun 2023 01:57 |
| Uncontrolled Keywords: | plant sensing system; estimating crop maturity; camera-based |
| URI: | https://sear.unisq.edu.au/id/eprint/40744 |
Actions (login required)
 |
Archive Repository Staff Only |
