Machine vision-based crop monitoring of capsicum grown in high-tech glasshouse

Abstract

The increasing demand for quality, year-round food production in limited space has led to the widespread adoption of protected cropping. Effectively monitoring and maintaining crops within these facilities requires substantial labour and expertise. Traditional manual monitoring is labour-intensive and time-consuming. Therefore, non-destructive image-based techniques, particularly those utilising RGB and 3D structural data, have gained attention. Protected crop environments are compact and complex; therefore, the involvement of computer vision researchers is required to solve specific problems in image-based crop monitoring, and the limited availability of such data hinders their involvement. In particular, monitoring of vertically supported tall crops such as capsicum has been underexplored in literature and is identified as a research gap. In addressing the gap, this research proposes a single RGBD camera and deep learning-based strategy for growth monitoring of vertically supported tall crops in state-of-the-art glasshouses. Deep learning and computer vision-based methodologies for plant height estimation (Chapter 4) and deep learning-based single leaf area estimation (Chapter 5) were developed as the primary growth monitoring methods. An extendable hardware platform was designed and developed for image data collection (Chapter 2). A time-series image dataset of a glasshouse capsicum crop was also produced and published using the proposed data collection strategy targeting computer vision application research (Chapter 3). The proposed height estimation methodology used Mask R-CNN for plant top and base segmentation; neighbour analysis was employed to increase plant top identification accuracy, clustering for noise filtering in 2D to 3D transformation of segmentations, and 3D processing to calculate plant heights. In conclusion, this research contributed to automated growth monitoring of the state-of-the-art, vertically supported glasshouse crops by proposing computer vision and deep learning-based plant height and individual leaf area estimation methodologies. Furthermore, it addressed the design and development of image data collection hardware and the provision of a dataset for computer science researchers. The promising results demonstrate the future of automated crop monitoring for better crop production in complex glasshouse setups and open new future research directions.

Date of Thesis submission	2024
Original language	English
Awarding Institution	Western Sydney University
Supervisor	Oula Ghannoum (Supervisor), Yi Guo (Supervisor) & Wen Hu (Supervisor)