Research On Automatic Detection Of Cotton Growth Stages By Image Processing Technology

Precision agriculture is a trend of current agricultural development, and extracting thecritical growth information of crop is the foundation for the implementation of precisionagriculture. The development of computer vision technology has helped a lot forautomatic monitoring the state of the crop growth. However, the current detection of cropgrowth using image processing techniques mainly concentrated in pest identification,weed identification, etc. And manual observation work has many disadvantages such aswork intensity and human error, so we want to apply the image processing techniques tothe automatic recognition of the crop growth state. In this paper, we regard the cotton fielddigital image as our research object, develop the automatic detection algorithms of cottonseedling stage, three true-leaves stage and five true-leaves stage.The automatic detection algorithm of cotton seedling stage is divided into two stages,the first step is segment seedlings, then extract crop rows and determine the seedling stagecoming date by the number of rows. The automatic detection algorithm of cotton threetrue-leaves is based on the crop coverage of the top view images. The coverage statisticsrequire the crop segmentation. Take into account the area of three true-leaves stage leavesis getting larger, the crop segmentation algorithm is more sensitive to light changes, weuse EASA algorithm to segment the crop pixels. Then we calculate and track the globalcoverage and local coverage. The automatic detection algorithm of five true-leavescombines coverage and the changes in plant morphology. Firstly, calculate the globalcoverage, then detect the single plant in front view images which including main stemdetection, side stem detection and node number. To detect the main stem, we segment thefront view images, then detect lines according to the edges and skeleton of segmentationresults. Side stem detection is according to the plant position which is determined by mainstem detection. Node numbers could be calculated by the position of main stems and side stems. In addition, we detect the algorithm validation by compare the manual observationrecords and experimental results of2011and2012cotton images.