Study On Vein Image Extraction And Three-Dimensional Modeling Of Flue-Cured Tobacco Leaves

Tobacco samples play an important role in guiding tobacco grading and classification during the processing.However,the physical samples are difficult to preserve and the sharing of information is also challenging,The traditional 2D photo recording of sample information also suffers from the problem of poor expressiveness.Therefore,researching how to establish a digital model of tobacco samples,including digital storage and 3D visualization,which can reflect the morphology and structural characteristics of physical samples related to grading as realistically as possible,is of great significance.Modeling tobacco,which is an irregular shape,has a key and challenging problem of modeling its veins,especially the blurred morphology of veins after roasting,which makes it more challenging to achieve lifelike modeling.Moreover,the traditional L-system plant modeling method based on fractal theory is usually based on simplified rules and parameters that generate relatively simple and abstract plant models,which cannot simulate the complex growth morphology of real plants and lack details and realism.Therefore,this paper proposes a modeling method for tobacco veins based on a combination of Cycle-Consistent Adversarial Networks(Cycle GAN)and L-systems.The specific research work is as follows:(1)Research on the extraction method of the veins of roasted tobacco leaves.According to the research ideas,it is necessary to extract the vein images from the actual roasted tobacco leaf images to provide training samples for the subsequent vein generation model.First,a "priority grayscale conversion method" is proposed to preprocess the tobacco leaf image,and the CMY and RGB values of the leaf color channels are adjusted in order to convert the tobacco leaf image to grayscale.Combined with image enhancement,it can better distinguish vein pixels from the background and leaf meat pixels.Then,compared with the traditional fuzzy C-means clustering(FCM)method for vein image segmentation,a new method based on density-based spatial clustering of applications with noise(DBSCAN)is proposed.Experimental results show that DBSCAN can effectively remove noise in the leaf image and extract clearer and more complete main and primary vein images.(2)Research on the 3D modeling of the veins of roasted tobacco leaves.A method that combines cycle consistency loss generative adversarial networks(Cycle GAN)and fractal theory is proposed to complete the vein images of tobacco leaves.First,multiple tobacco leaves vein images are extracted as samples for training Cycle GAN.Then,according to the learned generation rules,the first and second-level vein images are recursively generated.Next,based on the fine vein features of tobacco leaves belonging to the non-recursive fractal,the generation rules of the fine vein image are defined in the fractal theory L-system,and the fine vein image is generated.Finally,the generated vein images are imported into the vector conversion software Adobe Illustrator to complete the 3D modeling of the veins of roasted tobacco leaves.(3)Based on the theoretical research results of this paper on the extraction of the vein images of roasted tobacco leaves and the 3D modeling of the veins,an application prototype system was analyzed,designed,and developed according to the software engineering process to verify the feasibility of the proposed method and model.Theoretical research results show that the proposed model can effectively solve the problem of 3D modeling of the veins of roasted tobacco leaves,and the designed and developed application prototype system proves its feasibility for practical applications.