| The information about the growth stage of crops in the farmland is not only an important basic data for analyzing the relationship between the growth process of crops and different management measures, but also has been applied to many respects in precision agriculture, such as thinning, filling the gaps with seedlings, applying fertilizer, irrigation, pruning, harvest and insect disease prevention, etc. By far, the information is mainly acquired manually. However, this method is featured by many disadvantages, for instance, requiring abundant labor force, excessively relying on the operators’ subjective judgment, time-consuming and lacking of real-time property, etc. Therefore, it is necessary to propose a non-contacting, continuous and automatic data acquisition method.Featured by the obvious advantages and characteristics of high precision, fast speed and abundant information in terms of acquiring the information of crops, computer vision technology has huge potential in saving labor force and reducing the subjectivity of manual judgment. By far, it has been widely applied to monitoring the growth status of crops. However, there are few researches on applying digital cameras to continuous monitoring crops in the farmland during different growth stages.In this paper, the author automatically identifies and extracts the three key growth stages(emergence stage, three-leaf stage and four-leaf stage) of rapeseed plant by using computer vision technology. The main research contents and the achievements include the following four respects:(1) A computer vision-based image acquisition system for rapeseed plant is built. The composition about hardware and software of this system is discussed. The variety, planting place,planting methods and image acquisition modes for On the field of rapeseed plant are described(2) The rapeseed plants in the farmland images are segmented. The first task is to segment rapeseed plants from the image in order to identify different growth stages of rapeseed plant. In view of different interior and outdoor weather conditions as well as the influence from complicated environmental factors in the farmland, in this paper, a new method for segmenting crops(HI-hue intensity color segmentation of Gaussian distribution model)is researched. Firstly, the author acquires the image information(with the background removed) of rapeseed plants under different developmental stages to serve as training sample. Then the RGB color model of the sample data is converted into HSI color model. Then calculate the mean and variance of all the strength I corresponding hue(Hue) value in the HSI color model, and HI lookup table is built. Finally, every pixel is judged successively to see whether it belonged to the crop. In this paper, comparison is carried out on HI color segmentation method based on Gaussian distribution model and the four existing segmentation methods. Qualitative and quantitative evaluations are carried out on the experiment results after segmentation. The experiment results show that the error of the comprehensive segmentation precision proposed in this paper is 2.67%, and the standard deviation of its segmentation stability is 2.97%. It has the maximum performance compared with other methods. Moreover, it has robustness and is not sensitive to the change in indoor and light conditions as well as complicated environmental factors.(3) Emergence stage is initiatively identified by using shape parameter method. segmentation results of the field rapeseed plants are used, a series of morphological conduct rapeseed plants: hole supplement, open operation and Moore neighborhood profile tracing, etc., for the purpose of reducing the influence of other factors on the identification effect. Then the features of two typical geometrical shape(area value and density value) of the cotyledon during emergence period of rapeseed plant are calculated. Comparison is carried out on these values and the empirical values consistent, thus judging whether the rapeseed plant is in the emergence stage. In this research, two evaluation indexes of correct identification rate and wrong identification rate are proposed to evaluate the experiment results. The evaluation shows that the experiment results are consistent with the results of manual observation.(4) The rapeseed plants are judged by using active shape model(ASM) of pattern recognition to automatically recognizer whether they are in three-leaf stage or four-leaf stage. Firstly, the profile information of three-leaf stage or four-leaf stage of rapeseed plant is acquired by means of manual calibration, thus obtaining the geometrical shape information of the three-leaf stage or four-leaf stage of an entire plant. Model statistics calculation is carried out on the geometrical shape information of the sample data, through which the average shape models of three-leaf stage and four-leaf stage are obtained respectively. Texture model for local gray level is built. Then the weight factors affecting shape changing are constrained. Finally, the objectives similar to the shape models are matched in the image, so as to automatically identify three-leaf stage and four-leaf stage of Rapeseed plant. this paper uses single and multiple Rapeseed plant images to do the experiments,and are able to achieve automatic recognition of three-leaf stage and four-leaf stage of Rapeseed plant.Automatic identification method of Rapeseed plant growth and development based on vision technology that this paper is proposed is feasible, it can automatically identify emergence stage, three-leaf stage and four-leaf stage of Rapeseed plant.the research results can provide a basis for field management decisions, thereby serving the mechanized production of Rapeseed plant. |
PDF Full Text Request