Remote Sensing of Nutrient Deficiency in Lactuca sativa Using Neural Networks for Terrestrial and Advanced Life Support Applications
Document ID: 316
Doctoral Dissertation
1 The Pennsylvania State University, Department of Agricultural and Biological Engineering, University Park, PA
Abstract
A remote sensing study using reflectance and fluorescence spectra of hydroponically grown Lactuca sativa, ‘Ostinata’ (lettuce) canopies was conducted. An optical receiver was designed and constructed to interface with a commercial fiber optic spectrometer for data acquisition. Optical parameters were varied to determine the effects of field of view and distance to target on vegetation stress assessment over the complete growth cycle of the test plants. Feedforward backpropagation neural networks (NN) were implemented to predict the presence of canopy stress and vegetation nutrient status based on spectral inputs of canopy reflectance and fluorescence. Effects of spatial and spectral resolutions on stress predictions of the neural network were also examined.
Visual inspection and fresh mass values failed to differentiate between controls and plants cultivated with 25% of the recommended concentration of phosphorous (P) in solution. There was also difficulty discerning 25% nitrogen (N) from 5%P based on fresh mass and visual inspection. The NN’s were trained on input vectors created using reflectance and test day, fluorescence and test day, and reflectance, fluorescence, and test day. Four networks were created representing four levels of spectral resolution: the COLORBLOCK NN with wavelength resolution of Δλ≈100-nm, 10-nm NN (Δλ≈10-nm), 1-nm NN (Δλ≈1-nm), and 0.1-nm NN (Δλ≈0.1-nm).
Results from the NN validation classification demonstrated that all four types of network could be used as a remote sensing method for detecting extreme nitrogen deficiency early in the growth cycle. For the lower resolution models, COLORBLOCK and 10-nm NN, the best classification results of 5%N occurred using both reflectance and fluorescence spectra and a field of view (FOV) encompassing a 7.5-cm diameter spot size or 0.8 of a plant. For the higher resolution models, 1-nm NN and 0.1-nm NN, 5%N specimens were classified the best using reflectance spectra without fluorescence. For the highest resolution model, 0.1-nm NN, the best classification results coincided with the largest FOV, a 15-cm spot size or area encompassing 3 plants.
The 10-nm resolution was found to be sufficient for classifying extreme nitrogen deficiency in freestanding hydroponic lettuce. As a result of leaf angle and canopy structure broadband scattering intensity in the 700-nm to 1000-nm range was found to be the most useful portion of the spectrum in this study. More subtle effects of “greenness" and fluorescence emission were obscured by canopy structure and leaf orientation.
Scans having higher than acceptable variation should then be deleted from the neural network training and testing. This would enhance the robustness of the system. As field of view was not as found to be as significant as originally believed, systems implementing higher repetitions over more uniformly oriented, i.e. smaller, flatter, target areas would provide for more discernible neural network input vectors.
It is believed that this technique holds considerable promise for the early detection of extreme nitrogen deficiency in Lactuca Sativa cultivated in NASA Advanced Life Support well as in terrestrial hydroponic systems. Further research is recommended using stereoscopic digital cameras to quantify leaf area index, leaf shape, and leaf orientation as well as reflectance. Given the additional information provided by stereoscopic vision systems, fluorescence emission may also prove to be a useful biological assay of freestanding vegetation.
| Citation: | E. S. Sears, "Remote Sensing of Nutrient Deficiency in Lactuca sativa Using Neural Networks for Terrestrial and Advanced Life Support Applications", The Pennsylvania State University, Doctoral Dissertation, April 2001, 224 pages |