| Providing real-time and dynamic feedbacks for greenhouse control system is one of the key method by real-time, accurate and fast acquisition of plant growth information. On the purpose of improving availability of greenhouse environment control system, based on real plants growth conditions, considering relationship of plant growth morphologies and environments, a laser vision-based 3D scanning system on plants was set up in this thesis. Dynamic monitoring of plants growth in greenhouse and 3D morphological measurement of plants were implemented. On the base of 3D morphological parameters of plants, multi-scaled relationship of plant growth rate and its environment variation trend was analyzed by EEMD decomposition. On different time-scales, influence factor on plant growth rate was obtained by analyzing of plant growth rate on different time-scale. Multi-scaled model of plant growth rate and its environment variation was conducted. The optimal solution of envirionment combination of the model was obtained. The research was detailed below.(1) Establishment of laser vision-based 3D scanning system on plant growth monitoring and plant morphological parameters measurement. Patterns on the 3D calibration board was modified by adding donut shape to fast the calibration of camera parameters and laser plane functions, since there were more geometric features on donut shape pattern. In traditional way, central line of 3D calibration board was required to be projected in the middle of screen rigidly, which slowed down the calibration. To tackle this trick, a new method, based on priori-knowledge, was conducted to make a distinction between the left calibration board and right calibration board.(2) Extraction of laser stripe centroid at sub-pixel precision level based on the HIS color-space. In greenhouses, laser stripe image was affected by uneven sunlight with lots of noises and the errors of centroid extraction was severe. Given this, HIS color-space based segmentation was described in some detail. In the HIS color-space, component H and S were used to segment the laser strip image to locate the laser stripe in the image. In the experimental images, there were more than one laser stripe on a column because of different surface depth information on the surface of plants. Every strip was marked first. Threshold was defined by the average gradient of laser strip, and adaptive centroid extraction was implemented.(3) Registration of plant point cloud data at multi-view based on mutual corresponding estimation.3D point cloud data from single view could not express the whole status of plant growth, especially for plants with plenty of branches and thick leaves. Multi-view data acquisition was necessary. Based on point features histogram, features in two data collections was reached by means of kd-tree nearest neighbor query. And then, corresponding points in two data collections was determined. Corresponding estimation was conducted by mutual corresponding estimation. Last, random sample consensus was used to register two point cloud data sets. Experiments on 8819 Hot pepper was conducted. Pepper plant was scanned from 6 view point at 60 degree. And all point cloud data were registered into a whole model of the plant.(4) 3D skeletonization of plant point clouds using voxel thinning algorithm and improved Laplace operator algorithm. In order to provide a more efficient platform for plant morphological analysis and more storage space for huge plant point cloud model, skeletonization was an effective approach. Plant 3D point cloud model of water spinach and 8819 hot pepper were skeletonized by voxel thinning and improved Laplace operator respectively. Results of test experiment showed that performance of voxel thinning was less accurate on precision and slower on processing speed than that of improved Laplace operator.(5) Plant growth morphological parameter calculation based on local point feature operators. Different methods were used to calculate the plant morphological parameters including height and width of plants, leaf length, leaf area and fruit size. The methods were tested on water spinach and 8819 hot pepper respectively. A new iteration method was proposed to acquire central axis points of leaf point cloud data set. By means of B cubic spline interpolation, axis points was fitted and its length was calculated by differentiation method. Leaf point cloud data was triangulated by Delaunay Triangulation. The point cloud data was transferred to curved surface, composed of plenty of small Delaunay triangular. Area of all these Delaunay triangular constitute area of leaf. Size of fruit was tested on hot pepper fruit by the same method of leaf length calculation. The external cuboid surrounding plant point cloud data was used to compute the height and width of plants. All parameter method were used to acquire growth parameters and monitor growth status in growing period.(6) Decomposition and analysis of plant growth rate and environemt based on Ensembled Empirical Mode Decomposition. Decomposition of single water spinach total leaf area and environment elements(temperature, relative humidity, total inner radiation) in 2013 and 8819 pepper plant height and environment elements(temperature, relative humidity, total inner radiation) in 2014 were conducted in the form of IMFs in this dissertation. Peroid oscillations of IMFs were calculated. According to the period oscillation, IMFs were catagoried into hour-scale and day-scale. Analysis of plant growth rate on different time-scale presented that plant growth rate on day-scale was influenced by nutrient solution applying, while plant growth rate on hour-scale was influenced by instant frequency of IMFs of environment elements. Multi-scale analysis results was followed.a) On day-scale,leaf area growth rate on day-scale increased greatly after nutrient solution applying and maximummed in 2-4 days, while pepper plant height growth rate increased and maximumed in 3-5 days.b) On hour-scale, both leaf area growth rate and pepper plant height growth rate were affected by instant frequency of envirionment IMFs’oscillation. The two experiments were finished covering the whole rainy season in Nanjing. Single water spinach total leaf area growth rate increased as the temperature instant frequency increasing and raletive humidity instant frequency decreasing, accompanied by a certain time delay in the whole experiment time, while it changed on an opposite trend with that of instant frequency of environment element IMFs in the rainy season and on the same trend after the rainy season.8819 pepper plant height growth rate changed on the same trend with that of instant frequency of environment IMFs before the rainy season and on an opposite trend in the rainy season.(7) Correlation models of plant growth rate and environment instant frequency on hour-scale was established, and the optimal solution of environment elements combination was soluted by means of nonlinear multivariate constrained optimization method. Taking the examples of single water spinach total leaf area and 8819 pepper plant height, the correlation model was built by polynomial fitting with environment instant frequency on hour-scale as features. Both correlation models were regarded as target funcions, and nonlinear multivariate constrained optimization method was used to solve the optimal solutions of the two target functions. Optimal combinations of environment elements instant frequency were obtained at the value of temperature 1.07, raletive humidity 6.85, total inner radiation 0.5 for single water spinach total leaf area and temperature 0.6, raletive humidity 0.2, total inner radiation 4 for 8819 pepper plant height.
|
PDF Full Text Request