On the other hand, many advanced machine learning techniques extract “factors” or “features” from the data that are not “things” with a “physical” reality 13. However, since stress reactions result from a complex web of interactions between the genotype and the environment 12, common data analysis methods for plant phenotyping such as spectral vegetation indices run the risk of leading to an over-simplified or even misleading interpretation of spectral responses to stress as they consider only few distinct wavelengths. This indicates that phenotyping processes can benefit from hyperspectral data analysis and machine learning techniques which can uncover the characteristics of how plants respond to environmental stress. The work presented here is motivated by the insight that hyperspectral measurements can reveal relationship between the spectral reflectance properties of plants, and their structural characteristics and pigment concentrations, which are considerably influenced by biotic plant stress 3, 11. The reflectance values of continuous wavebands of the electromagnetic spectrum are influenced by various plant characteristics any kind of stress causes complex changes in the plants’ physiology and composition which, in turn, alters the spectral reflectance pattern (=spectral signature) of plants in the visible range (VIS, 400–700 nm), near-infrared (NIR, 700–1000 nm) and short wave-infrared (SWIR, 1000–2500 nm). Especially hyperspectral imaging data of individual plants or crop stands contains an enormous amount of information on their physiological and biochemical status 7, 9, 10. This kind of sensor-based phenotyping has already been proven successfully for monitoring physiological traits and plant genotype-specific responses to biotic and abiotic stresses 6, 7, 8. Each of these diseases causes characteristic symptoms and the need to improve and to automatize their monitoring in fields and/or greenhouses has led to an increasing adoption of technologies such as hyperspectral imaging. Barley, for example, may be affected by various foliar pathogens during the vegetation period, and significant quantitative and qualitative yield losses are caused by diseases like powdery mildew, net blotch and brown rust 5. Unfortunately, these scale badly to the growing amounts of data in plant phenotyping and are prone to human conformation bias. Today’s approaches to disease detection and planning of plant protection measures still very much rely on human experts and/or on prognosis models. Especially the detection of plant diseases is an important task in crop production to avoid yield losses, and in plant breeding for the selection of diseases resistant genotypes.
Within this context, non-invasive sensors and computer based technologies demonstrated their potential to equip todays agriculture with tools to solve current and future challenges 4. Recently, phenotyping is defined as a set of methodologies and protocols to assess plant parameters at different scales 2, 3. The plant phenotype is of importance to evaluate the performance of a crop as the interaction between a plant genotype and its environment 1. In short, our analysis and visualization of characteristic topics found during symptom development and disease progress reveal the hyperspectral language of plant diseases. We also present a visualization of the topics that provides plant scientists an intuitive tool for hyperspectral imaging. Based on recent regularized topic models, we demonstrate that one can track automatically the development of three foliar diseases of barley. Then, we apply probabilistic topic models, a well-established natural language processing technique that identifies content and topics of documents. To uncover latent hyperspectral characteristics of diseased plants reliably and in an easy-to-understand way, we “wordify” the hyperspectral images, i.e., we turn the images into a corpus of text documents. In this work, we present an approach to plant phenotyping that integrates non-invasive sensors, computer vision, as well as data mining techniques and allows for monitoring how plants respond to stress. In particular, hyperspectral imaging have been found to reveal physiological and structural characteristics in plants and to allow for tracking physiological dynamics due to environmental effects. import java.io.Modern phenotyping and plant disease detection methods, based on optical sensors and information technology, provide promising approaches to plant research and precision farming.
0 kommentar(er)
