Automated machine vision sensing of plant structural parameters

Abstract

[Abstract]: Automated sensing of crop water stress is required to provide real-time variable-rate irrigation control that responds to spatial and temporal variability of plant water needs. Plant structural parameters such as internode length in cotton (i.e. the distance between successive main stem branches) are recognised as significant indicators of water stress level. This paper demonstrates the successful automatic identification of nodes and measurement of internode lengths. A moving in-field camera enclosure involving plant contact was trialed in the (Australian) 2005/06 cotton growing season. The enclosure continuously traversed the crop canopy, collecting video footage of the cotton plants. The enclosure design utilises the natural flexibility of the growing crop to force (non-destructively) the plant’s main stem onto a fixed object plane which enables the direct measurement of geometric dimensions without the need for stereo vision. Plant features are discriminated via processing of successive images to identify the main stem and branches, and hence stem-branch junctions (‘nodes’) are located. After confirmation that the plant structure is in the required object plane (by comparison of adjacent frames), the measurements of internode lengths are calculated. From fourteen sequences of images, with typically fifty images per sequence, main stem identification has been achieved in up to 88% of frames, and internode lengths have been measured with standard errors of 6% via automatic image processing and 3% via manual identification of nodes. It is also demonstrated by analysis of the computational requirements that the necessary image processing can be undertaken in real-time. It is therefore concluded that on-the-go measurement of plant structural parameters is feasible and may be implemented at relatively low cost.