Reasearch On Method Of Identifying Weed In Corn Seedling Field Applying Image Processing Technology

At present the extensive method of spraying the herbicide not only pollutes air butalso damages the quality of field and increases the costs of agriculture production. In thissubject, we study the method to identify the inter-row weeds from corn seeding.Accordingly the method can provide theory and technology support forvariable-controlling spraying method in corn field and decreasing dosage of herbicide.The main content and achievements were presented as follows:1. Pre-process the original image to remove the noise. In this subject comparing 3filter algorithms during research (lowpass filter, neighborhood filtering, median filtering),and applied neighborhood filtering to filter original images.2. Applying color feature increased contrast changed between the green plants andsoil, separated the plants from complex background and color image to gray-level image.3. Some algorithms of image threshold segmentation were researched and selected asuitable segmentation method, namely oust threshold segmentation, the algorithms couldtransfer gray-level image into binary image.4. Morphology filter are used to erase the random noise in binary image5. Study the image features include geometric characteristics, nondimensional shapefeatures, regional characterized moment. The results of the statistical analysis show thatthe geometric characteristics of corn significantly greater than other weeds. And a singledimensionless parameters or regional special moment can not effectively distinguish theweeds. Area can be considered as the most effective feature for identifying, and thencombined it with other shapes of a feature set to identify weeds.6. A back-propagation neuron network was designed for weed identification.According to the experiments, the optimal network structure was 4-5-2 with the traininggoal 0.03, training speed 0.3. test results show that its small quinoa, iron amaranth,buckwheat dish, crabgrass recognition rate: 96%, 92%, 96%, 94%.7. Software for weed identification was developed by use of Visual C++. It canbeused in image processing. It can also provide technology support for developing weedidentification system in future.