The Method And Experiment Of Mountain Tobacco Population Statistics Based On UAV Images

In the process of tobacco planting,it is necessary to make accurate statistics of the planting area and the amount of planted tobacco,so as to facilitate the planned purchase in the later stage.At present,the tobacco leaf production management department is usually use manual counting methods to count the number of tobacco plants in the field at the beginning of tobacco seedling transplantation.This method is time-consuming,laborious,inefficient,and prone to errors.The landforms of Guizhou belong to the plateau mountains in southwestern China.The terrain is rugged and mostly slopes.The method of manual inventory is more difficult.In order to reduce the manual labor intensity,improve work efficiency and data accuracy,this paper combines information technology with agricultural technology,and proposes a statistical method for the number of mountain tobacco plants based on drone images.This method can meet the statistical requirements of the number of tobacco plants in mountainous areas and has important practical significance for the management of tobacco market.The main research contents of this paper are as follows:1.The collection device and method of tobacco image are studied.The influence of weather,camera error and topographic relief on aerial image are analyzed.The collected images are preprocessed with clipping,distortion correction,grayscale and smoothing enhancement,which prepared for the study of real-time tobacco image recognition in the later stage.2.The photos of the tobacco seedlings in the mountain,which are collected by unmanned aerial vehicle(UAV),are taken as the research object.A fast stitching method based on SIFT algorithm is proposed.This method effectively reduces the number of matching feature points.Compared with the traditional SIFT algorithm,the image of the tobacco collected by the drone is 49.8%faster than the traditional SIFT algorithm,which effectively saves the image stitching time and significantly improves the image stitching efficiency.3.The relationship between illumination intensity sensitivity and color components in RGB color space and HSI color space is analyzed,which provides a basis for color space selection in color segmentation.In the later image identification of tobacco plants,the phenomenon of misjudgment of tobacco plants and weeds was studied in the field of3~8 leaf seedlings in Guizhou mountainous areas.A field applicability identification method with high applicability is proposed based on the combination of theoretical analysis and case test.4.Due to the different flying height,light intensity and terrain gradient of the UAV in image acquisition,the accuracy of image processing technology is affected in the later stage of tobacco plant quantity identification.This paper analyzes the influence of these factors and proposes solutions to improve the accuracy of tobacco plant quantity identification.5.In order to obtain the optimal period of tobacco plant recognition,the tobacco plant images in the same region of 3~8 leaf period collected are processed and analyzed under the same conditions.Then the data obtained from the processing and analysis are fitted linearly.It can be seen from the fitting results that the larger the correlation coefficient R~2,the better the recognition effect.6.Based on MATLAB GUI platform,using the method given in this article and the importing existing programs,the automatic statistical software for the number of images of tobacco plants is written and developed.Then the tobacco area is processed and analyzed.The results show that the application of software to the detection of the number of tobacco plants in the mountain can significantly improve the inventory speed and recognition accuracy,and has a high field adaptability.