Study Of The Plant Information Acquisition And Processing Technology Based On Vision And Image

Machine vision and image processing technology are widely used in the diagnosis of plant growth status, growth prediction model, disease testing, quality inspection and grading, precise control and other field of agricultural engineering. In this paper research the algorithms of collection and processing of plant information by image processing, hyperspectral, LiDAR and embedded machine vision technology. By the leaf area measurement, sugarcane node identification and location, canopy projection area and volume measurement and fruit volume and surface area measurement to verify the reliability of the algorithm and platforms, plant organ (leaf area) measurement has guiding significance for growth status diagnosis and growth prediction model, acquiring plant information (sugarcane node) is important for automated planting, acquiring vegetation(forest) group information can be used to efficiently manage and scientific decision-making. Research in these areas from the micro to the macro, from organs (leaves) to plants (sugarcane), to the population (a large range of trees), and finally studied specific application of information acquisition and processing based on the Android platform, they are all presented based on information acquisition technology, feature information extraction technology and information analysis technology.In this paper, the main conclusions are as follows:(1) A novel method to establish three-dimensional model by multi-angle images have been proposed in this paper in order to measure the surface area of leaf in natural grow state. Shot from multi-angle to get the images of planar calibration grid, then get the camera parameters by calibration algorithm, reduced the image distortion, improve the accuracy of the modeling measurement.Shot from various angles of leaf in natural grow state, and processed images by Photomodeler software, got three-dimensional point cloud of leaf; then using MATLAB programming to get the three-dimensional surface modeling and computing the surface area of leaf; combinatint the scanner and Photoshop software to measure the leaf, comparred this method with the previous method finally. The experimental results show that the proposed method is a good method of measurement leaf in natural grow state, it has a high-precision up to 99%.(2) Achieved the identification and location of sugarcane node by hyperspectral imaging technology. The color of sugarcane internodes and nodes are similar to each other, and with the interference of white fruit powder on the skin, that affect node recognition and location seriously, this paper proposes a method using hyperspectral imaging for node identification and localization. In this study,236 sugarcane samples were collected by the hyperspectral imaging acquisition system, then extracted the average spectrum of region of interest, used successive projections algorithm (SPA) to extract characteristics band (1022 nm,1062 nm,1456 nn,1609 nm and 1649 nm), established the PLS discriminant model by these 5 characteristic band, the recognition accuracy rate of calibration set and prediction set were 99.44% and 98.31%, the performance of the model was good. Calculation the sum of 5 characteristic band spectral data, the output results were classified by the threshold of 0.5, got the binary image of sugarcane, used image processing for quantitative analysis, statistic the sum of each row of pixel values, found the maximum value to locate the position of node. Finally, analyzed artificial and image measurement results of node position, the standard deviation was 0.7 mm, the maximum absolute error was 2.6 mm, the results had high reliability, meet the actual production requirements. The results indicate that the hyperspectral imaging technology combine with SPA-PLS method can identify and locate the sugarcane node, provide the theory method and basis for the research of intelligent cutting instrument of sugarcane which can prevent injury buds.(3) Achieved the extraction of projection area and volume of a wide range of trees. Airborne LiDAR can obtain three-dimensional structural information of trees, get forest information by using points cloud classification and calculation method of crown projection area and volume. First, the points cloud of ground were extracted by triangulated irregular network, then the points cloud of building were extracted by using planar-fitting filtering algorithm. Filter points cloud of building to get points cloud of forest, and the points cloud of forest were projected onto x-y plane. The edges of points cloud of timber were extracted by using the angle method; its corresponding image was display; the area of polygon were calculated by using area algorithm; then the calculation of the projection area of crown was achieved, combine the formula to calculate crown volume. In the study area,10 experimental districts were randomly selected to carry out the traditional manual measurement. The experimental results showed that the projection area and volume's correlation coefficient of the two measurement methods were 0.957 and 0.944, respectively. The proposed method was feasible to achieve the accurate extraction of forest information that extracts a wide range of tree crown projection area and crown volume rapidly and efficiently.(4) Achieved quick and nondestructive measurement of the volume and surface area of irregular shape of agricultural products, used of Java combined with OpenCV library to develop software, which taken Android Tablet PC as platforms, using image processing techniques to measure the volume and surface area. The measurement steps were as follow:image acquisition, image segmentation, image binarization, filtering denoising, calculation the coordinate of three-dimensional wireframe model, volume and surface area calculations. Used this method to measure the circle having a diameter of 70,90,110,130 and 150 mm at six distance that between 100 mm to 350 mm, and the segment spacing was 50 mm, verified that the camera distortion was 0.3%, the contour size of the measured object had scarcely effect on the measurement accuracy. Application of the proposed method to measure the volume and surface area of 10 citrus, and compared with the measurement results of the drainage method and scanning method, accuracy rate was more than 98%. Test the number of images that ranged from 5 to 15, measurement accuracy increases with the number of images, calculation time were within 5s. Experimental results showed that the proposed method has the ability to obtain accurate measurements of volume and surface area.