The Research Of Statistic Iteration Algorithm For Farm Field Image Segmentation

The traditional agriculture work is very tedious, low efficiency and occasionally dangerous, such as pesticide poisoning, sunburned skin or even worse. The emergency of agriculture navigation effectively resolves these problems. To detect path based on machine vision in automatic navigation is a main method, because of rich information, flexible use and not considered the circumstance (mainly based on the ridge, balk that is formed before in farmland and orchard). The image segmentation is the key factor in the vision detection of path. This paper mainly researches the statistic interation algorithm and its application in agriculture image segmentation. On the other hand, the auto navigation of agriculture vehicle meets the demand of the development of precision agriculture, and protects the circumstance.The image segmentation algorithm that used in agriculture navigation before couldn't effectively remove the imfluence of broken ridge, weeds, shadow and illumination change. So the followed guidance line detection algorithm must resolve these problems, it would result in lower robust because of the complexicity of itself, such as Hough thansform. The purpose of agriculture image segmentation is to detect the crop row, so the statistic iteration algorithm Meanshift and Support Vector Machine was used in this paper. In order to save running time the two statistic iteration algorithm were integrated with wavelet multi resolution individually. The color models that usually used in agriculture image segmentation were discussed from the point of theory view. In the last some typical agriculture image segmentation algorithms were contrasted with the algorithm that used in this paper based on experiments.The main innovation work of this thesis as follows:1. Based on the structure features of agriculture field, gives a new idea that only used the lower resolution image. This not only could remain the information that would be detected, but also effectively removed the influence of broken ridge, weeds and shadows, and the running time was saved too.2. Based on the former method and the features of farmland, this paper proposed a new fast Meanshift algorithm that divided the agriculture image into several pieces and run the Meanshift algorithm only on the seeds that selected from each pieces. Based on the wavelet multi resolution decomposion, the Meanshift algorithm remove the the influence of broken ridge, weeds and shadows further.3. Gave a rectangle model with weight coefficients to calculate the average value. This model considered the distribution of the color in agriculture field, that is the color is more similar in the crop row direction. The use of this model can remove the influence of broken ridge and weeds.4. The combination of averge and standard deviation as a new input feature of Support Vector Machine, the result imply that not only obtained a good segmentation but the width of crop row remained too. This is benefit for the followed navigation line detection.5. The Support Vector Machine algorithm were used in the agriculture image segmentation for the first time, and can removed the influence of broken ridge, weeds, shadow and illumination change.