Image-based analysis of fungal-damaged soybeans.
详细信息
- 作者：Ahmad ; Irfan Saleem.
- 学历：Doctor
- 年：1997
- 导师：Reid, John F.
- 毕业院校：University of Illinois
- 专业：Engineering, Agricultural.;Agriculture, Plant Pathology.;Computer Science.
- ISBN：0591471434
- CBH：9737029
- Country：USA
- 语种：English
- FileSize：10391574
- Pages：265
文摘
Image-based characterization and analysis of soybean seeds was carried out. Methods for characterization of fungal, viral, immature, and asymptomatic seeds were developed, which can result in economic savings by measurement of seed quality with a possible reduction in inspection. The study was successful in developing a knowledge domain based on color, morphological, and textural features for future development of intelligent automated systems.;bhe color analysis showed that there were distinct color differences between asymptomatic and symptomatic seeds. Classification accuracies were, asymptomatic 97%, Cercospora spp. 83%, green immature seeds 91%, soybean mosaic virus: black 81% and brown: 87%, Alternaria spp. 30%, Fusarium spp. 62%, and Phomopsis spp. 45%. The classifier performance was independent of the selection year of the seed sample.;Morphological analysis was predicated on the hypothesis that classification of asymptomatic and symptomatic seeds can be conducted based on morphological features. The results obtained supported this hypothesis. A model comprising six morphological features was identified for describing seed shape.;The overall classification accuracy for seeds with the highest probability of occurrence ranged from 75 to 88%. An appropriate set of invariant textural features were determined. It was observed that Fusarium spp. and asymptomatic seed were detectable from other damage types as hypothesized. Asymptomatic and symptomatic seeds with the highest probability of occurrence had a classification accuracy of over 95%.;The development of the multi-feature classifier was the culmination of color, morphology, and texture feature sets into a unique combination with optimum theoretic performance. At the individual seed level asymptomatic seed was shown to have 100% accuracy, with 90% for Cercospora spp.;Each of the three feature sets, color, morphology, and texture were able to discriminate specific seeds with varying degrees of success. A neuro-fuzzy inference system was developed to classify asymptomatic, Cercospora spp., and Fusarium spp. The classification accuracy for asymptomatic seed was 91.6%, Cercospora spp. 68%, and Fusarium spp. 95%. A multimedia computer-based soybean visual information and grading system was developed. The research concluded that fungal-damaged soybean seeds can be characterized based on their images.