Extracting And Enhancing Weak Information Of Heavy Metal Contamination Stress On Crops From Hyperspectral Data

Being one of the main eco-environmental issues, heavy metal contamination stress on crops seriously threats not only agricultural production and food security, but also human survival and the global environmental quality, which has attracted increasing attention recently. To monitor pollution in crops and farmland at large scale by remote sensing technology is regarded as one of the most urgent and practical problems to be solved. However, the heavy metal content of agricultural soil is generally tiny in natural environment, thereby, the spectral response of crops is weak and unstable. Moreover, the source of spectral variation hardly distinguishes from other environmental factors, such as coupling water and fertilizer, solar illumination, atmosphere etc. On the other hand, the response is so weak that there is no obvious representation in the hyper spectrum, and may easily be ignored. Therefore, to explore effective theories and methods for enhancing such tiny spectral information on heavy metal pollution stresses in large area farmland meets increasing demands on quantitative and fine remote sensing applications currently and, as a scientific problems, is necessary to be solved.The object of the study is to discriminate dynamically and measure accurately the weak and hidden information of heavy metal stress on crops in natural field eco-system. The maize and rice sample fields with different heavy metal pollution levels in Changchun and Jilin City were selected to carry out the hyper-spectral measurement for the key periods during growing season. The heavy metal content in soil and crops were analyzed in the laboratory. By the experiment validation, physical mechanism reasoning and mathematics modeling, the sensitive indices referring to the changes in chlorophyll, nutrient, water content of crops under heavy metal stress are studied. The more powerful indices are explored by wavelet analysis and dynamic fuzzy neural network, thereafter a comprehensive evaluation model for crops heavy metal stress is proposed in order to find out the tiny change using remote sensed data in large area. The main researches and conclusions of this dissertation are summarized as follows:1. Relationships between soil-crops feature index and hyperspectral data(1) Based on the experimental data, the distribution of heavy metal in soils and crops is analyzed. According to the change mechanism of chlorophyll, nitrogen and water content in crops under heavy metal stress, the ecological damage synthesized index can be used to evaluate the stress level more accurately than the complex pollution index.(2) By investigating the response of spectrum to physiological reaction of crops under heavy metal contamination stress, the spectral retrieval models of chlorophyll, nitrogen and water content are constructed. It is found that many indices including NDVI, MCARI, OSAVI etc. lose efficacy when heavy metal pollution exists, and this can act as an indicator. On the basis of relationship among spectrum, crop physiological indices and heavy metal stress level, a new index called'SIr'is developed which is then validated as a new efficient index for exploring the heavy metal stress.2. Early diagnose of heavy metal stress on crops(1) Combining the reaction of chlorophyll, nitrogen and water content under contamination stress and its spectral feature, the multi-dimension spectral feature space are constructed using the following indices , namely, SDg/SDr,FD933 and WI3 for rice and X23, NI15*NI17 and D1025 for maize to evaluate the stress level. Because the three indices are very sensitive to chlorophyll, nitrogen and water content variations respectively, such diagnose model has a better performance for indentifying the contamination stress.(2) The whole spectrum characteristics are revealed by spectrum binary encoding. It is found that the sum of binary codes in 680-720nm region indicates well the existences of heavy metal stress on crops, especially for rice.(3) The wavelet analysis method is applied to examine the details and abnormal of cropsspectrum. The decomposition using different mother functions are compared and db3 is selected to reconstruct the spectrum of crops. It is found that the wavelet coefficient in 700-750nm express an obvious different in crop spectral feature with different stress level. The correlation analysis of abnormal extreme values, amplitude and heavy metal stress level shows that the abnormal extreme values in non-polluted fields are less than 0.01, whereas the above abnormal parameters increase with increasing contamination stress. Wavelet analysis extracts the detail information of the spectrum, and easily find out the spectral feature abnormal.3. Pollution stress level evaluation modelIntegrating the fuzzy theory and artificial neutral net technology,a dynamic fuzzy neutral net (DFNN) model is construed to evaluate the stress level . The model makes use of the presentation ability for fuzzy information, and the self-study and non-linear computation ability of ANN. The model takes sensitive spectral index as the input layer; the output is heavy metal stress level. It is tested by much dataset that the DFNN model has strong classification accuracy, high efficiency and stability. Using this model, the heavy metal stress level can be evaluated rapidly and accurately.In summery, the study makes full use the spectrum information completely ranging from 350 to 2500nm in this thesis. Integrating the three indices representing the chlorophyll, nitrogen and water content respectively, and combining the overall feature and detail information of the crops reflectance spectrum, the model for stress level evaluation is proposed based on the multi-criteria theory. The dynamic fuzzy neutral net works efficiently and stably and the stress level information can be extracted rapidly. The weak suitability and strong instability of previous methods are overcome by the DFNN.