School of Agriculture, Food and Ecosystem Sciences - Theses
Permanent URI for this collection
Search Results
1 - 1 of 1
-
ItemRapid screening methods for superior trait selections in lentil and field pea breeding( 2020)Most of the lentil and field pea grown in Australia are exported to India and surrounding countries, to the Middle East, Turkey and North Africa. While each country may utilise pulses differently, common to all, is that quality is based on the visual characteristics of the whole-grain and split-pulse, and its cooking quality. One of the objectives of pulse-breeding programs is to ensure that the quality traits of new varieties align with the preferences defined by the export markets. Pulse-quality traits were historically determined using empirical tests to quantify seed size, colour, contamination and defects. Since many of these tests are time-consuming to perform, comprehensive quality evaluation is reserved for advanced germplasm. Therefore, adoption of rapid and objective methods would improve efficiency and consistency of quality evaluation and enable comprehensive assessment of early generation lines. Technological advances in digital imaging and machine learning has seen a broad application of machine vision to assess agricultural products. While there is extensive research in this field, there are still relatively few machine vision methods which have been developed for the quality-assessment of lentil and field pea grains. Within this study, rapid and objective methods were developed to assess three grain-traits, which related to visual characteristics of lentil and field pea and were identified to be important within breeding programs. The targeted applications were the classification of broad market classes of field pea, quantitation of bleaching discoloration within the ‘green pea’ market class and classification of split and dehulled fractions of lentil and field pea post milling. Machine vision algorithms were developed based on the analysis of multispectral images. Linear discriminant analysis, based on image-derived colour, shape and size features, was used for the classification of field pea market classes. The model was applied to sound and defective grain samples, achieving perfect classification of sound grain and distinguishing sound from defective grain with 97% accuracy. The extent of bleaching in green field pea samples was quantified through an objective model which was developed on visible reflectance spectra (spectrophotometric analysis) and subsequently adapted for image-based analysis of grain colour. The image-derived colour scores closely matched the spectrophotometric analysis and additionally enabled the distribution or uniformity of bleaching to be objectively quantified within each sample. Furthermore, through the image analysis scoring system, the relative susceptibility to bleaching, of each genotype, was also quantified. Milled fractions of lentil and field pea were classified through the application of artificial neural networks, where network architectures and inputs were compared. A convolutional neural network, trained on image-derived feature distributions, was found to be the most accurate and computationally efficient model. Machine vision is an expanding field of research which offers the potential for consistent, accurate and rapid product-quality evaluation. The results of this study demonstrate the efficacy of machine vision applications throughout the pulse value chain and particularly within germplasm enhancement programs. Adoption of machine vision systems can increase the capacity for comprehensive screening at all stages of breeding which is currently not practicable through standard assessment methods.