Automation of Luminescence Quantitation for High-Throughput Plant Phenotyping Using Image Processing and Deep Learning
dc.contributor.advisor | Valles, Damian | |
dc.contributor.author | Rahman, Mahbubur | |
dc.contributor.committeeMember | Stapleton, William A. | |
dc.contributor.committeeMember | Kang, Hong-Gu | |
dc.date.accessioned | 2024-01-02T16:49:56Z | |
dc.date.available | 2024-01-02T16:49:56Z | |
dc.date.issued | 2021-12 | |
dc.description.abstract | The plant’s innate immune system consists of multiple layers of resistance mechanisms including one conferred by resistance (R) proteins [1]. Virulent pathogens have found ways to evade and actively suppress the host’s defense responses and have acquired the ability to cause plant diseases [1]. A luminescence reporter strain of Pseudomonas syringae, a bacterial pathogen, was developed to expedite this quantitation process, which is critical in resistance trait analysis. To further facilitate the high-throughput pipeline of the plant resistance analysis, two algorithms are developed to automatically quantify the luminescence from infected plants’ images with the reporter strain captured in the dark condition. Additionally, low light images of the corresponding plant were captured to perform this computation effectively. Conventionally, the computation of luminescence in image-based plant phenotyping is a prolonged and tiresome process. It requires monotonous human labor to delineate targeted area and store the results. Two parallel methodologies are proposed in this thesis to compute the luminescence automatically. An image segmentation technique with tuned parameters identifies the mask of the leaves from low light images. Every image mask associated with the corresponding luminescence image computes the mean luminescence of the targeted leaf. Secondly, UNet, an evolution of Convolutional Neural Network (CNN) is used to segment the leaf mask at equal resolution as the input image. The U-Net architecture has a shrinking or down-sampling path and expanding or up-sampling path. The down-sampling path shadows the conventional CNN and has repeated application of two convolutions. The up-sampling path consists of transposed convolution and provides high-resolution localized detection. U-Net architecture is convenient for semantic segmentation with a smaller number of datasets. My proposed architecture is trained with annotated low light images. The mask, which, accompanied by the luminescence image, calculates the mean luminescence. Two results from two parallel approaches are analyzed and compared. The mean luminescence obtained from the image processing technique and the deep learning technique is linearly correlated with the bacteria population in the affected plant. Both proposed algorithms are prompt and comprehensible and can be adopted for streamlining high-throughput plant phenotyping. | |
dc.description.department | Engineering | |
dc.format | Text | |
dc.format.extent | 88 pages | |
dc.format.medium | 1 file (.pdf) | |
dc.identifier.citation | Rahman, M. (2021). Automation of luminescence quantitation for high-throughput plant phenotyping using image processing and deep learning (Unpublished thesis). Texas State University, San Marcos, Texas. | |
dc.identifier.uri | https://hdl.handle.net/10877/17848 | |
dc.language.iso | en | |
dc.subject | luminescence computation | |
dc.subject | image processing | |
dc.subject | deep learning | |
dc.subject | artificial intelligence | |
dc.subject | plant | |
dc.title | Automation of Luminescence Quantitation for High-Throughput Plant Phenotyping Using Image Processing and Deep Learning | |
dc.type | Thesis | |
thesis.degree.department | Engineering | |
thesis.degree.discipline | Engineering | |
thesis.degree.grantor | Texas State University | |
thesis.degree.level | Masters | |
thesis.degree.name | Master of Science |