Computer Vision Technology Based Research On Digital System Of Forage And Grassland

Grassland is the refreshable natural resource for human beings, and the dynamic monitor and digitalized government of grassland is vital for the sustainable strategy. At present, the research on the grassland digitalization is rare. Hence, it is of great importance in theory and reality to realize grass recognition by computer vision, which will make contribution to improve the accuracy of grass auto recognition, and data acquirisiton.Computer vision is a multi-disciplinary technology that takes advantage of computer, image acquisition system to imitate the human vision in converting image into digital signal. Different from the remote sensing based digital system of the grassland, computer vision emphasized on digital images of grassland and forage captured by digital cameras. Taking advantage of computer vision, the research focused on instrinsic feature extraction, so as to realize the functions such as auto recognition of forage, and microscope images mosaic, etc. The computer vision based digital platform of grassland and forage was developed by integrating the above function modules. The main contributions are listed as follows:1) Forage recognition system based on seed images was developed. Owing to feature extraction is the key module of the system, two algorithms based on Gabor wavelets and Local Preserving projections (LPP) and local fractal dimensions were proposed, respectively. In the frontier algorithm, Gabor wavelet was used for image decomposition to eliminate the effect of illumination variations and scale variations. The dimensionality of Gabor space was reduced effectively by LPP, and meanwhile the intrinsic manifold structure was kept. In the latter algorithm, the fractal dimension is an effective tool to describe the roughness of texture. The experimental results conducted on a seed image containing700seed images of14species,5categories demonstrated the effectiveness of the algorithm, with an average correct classification rate of98.57%and99.43%.2) Forage recognition algorithms based on leaves and plants were realized, prespectively. In the leaf_based algorithm, Blocked local binary pattern (LBP) was used for textural feature extraction, and ant colony algorithm was adopted to optimize the extracted features in feature matching. The top classification accuracy was90%experimenting on180images of18categories. In the plant_based forage recognition, the Gabor global features were extracted for classification, and then global features were calculated to category the forage. The top recognition accauracy were100%when experimenting15images of3categories.3) The images of parasite eggs can be used for image segmentation, auto classification and counting. Erosion, dialition, and watershed algorithm in Mathematical morphology preprocessed the egg images for the acquisition of egg contour. A certain threshold was set when counting eggs, and the distance between two centers of eggs less than the thresohold would be combined into one egg to reduce the effect of oversegmentation.4) Microscopic images of forage seed were stitched into panorama. SIFT can extract feature sets, defining descriptor and obtain initial matching points. RANSAC was used to eliminate incorrect matching points, and fade-in and fade-out were adopted for smoothing the seam, and Gamma correction weakens the effect of image quality degraded by illumination. Thus the obtained image has a better overview.5) Realization of Digital Grassland Forum based on computer vision technology. The basic function modules of the platform were developed by the tools, such as JSP, Mysql and Apache. The emphasis of the platform was the integration of the image mosaic and image recognition, realized by the combined programming of Java and Matlab. The special function is convient for the users to upload the images captured in the practical applications to obtain the corresponding species information.