Measurement method for yield component traits of maize based on machine vision

Abstract: The maize variety test is an important link in the process of crop genetic breeding. The different maize varieties will produce a large number of varieties phenotype data, which need to be collected, collated, recorded, statistically analyzed and stored. Some phenotype data are related to the maize yield, such as bald rate, ear rows, row grains and so on. These maize characters are often collected by the traditional manual measurement at present. For example, the ear rows can be calculated by the maize section image which destroys the maize to be tested .Another measurement method for the ear rows is to rotate and scan the maize, which is very difficult to meet the needs of high throughput maize variety test. Aiming at the above problems, the calculation model according to the color and biological features of maize has been constructed based on the machine vision technology in this paper. The calculation model can compute the maize character parameters precisely, such as bald rate, ear rows, row grains and so on. The experimental results show that the calculation measurement has the high recognition precision and speed. The ear length ,ear diameter ,ear rows ,row grains and other yield components are taken as example for verifying the above calculation model in this paper. The experimental environment settings for image acquisition model are as follows: non wide-angle CMOS pinhole camera (portable, low cast), acquisition environment of soft light and bright place (no special light source set). The camera is 5 million pixels, and the image resolution is 2942 pixels ? 1944 pixels. Shoot height is 55 cm, the shooting format is to A3. The algorithm is tested by the PC machine which is configured as a dual core cpu (1.9 GHz) and 2 GB ram. The method presented in this paper can overcome these disadvantages of traditional manual measurement, such as low efficiency, subjective error, and unable to retain the integrity of the original maize material. The method presented in this paper can fetch the parameters of 3D phenotypic traits based on the 2D visible light imaging, and separately establish the projection correction model, the rapid estimation model and the calculation model for ear rows and row grains .The method presented in this paper can calculate the ear length, ear diameter (the calculation accuracy can reach more than 97 percent), ear rows and row grains of maize accurately. The zero error rates of ear rows measured by the method presented in this paper can reach 93 percent, and the absolute error of row grains measured by the method presented in this paper is about 2 grains. In this paper, the acquisition speed for the maize characters also has been tested, and the experimental results show that the measurement speed is up to 30 per minute above. With the promotion of the PC machine parameters, the measuring speed can be greatly improved. The methods proposed in this paper have important reference value to achieve the high throughput test and fine breeding.