| Scientific and accurate projections of crop yields are of great significance in guiding food production,distribution and consumption,ensuring national food security and maintaining social stability.Accurate measurement of current wheat yield-related parameters in the field can not only check whether wheat production management is sound but also analyze the reasons for the increase or decrease in yields.The number of wheat ears and wheat yield per unit area are two important parameters for wheat yield measurement in the field.At present,the wheat yield information is mainly measured by adopting the field sampling survey method,which needs a lot of manpower and is very time-consuming and labor-consuming.In order to obtain the information related to wheat yield per unit area in field quickly and accurately,and solve the problems of low automation,time-consuming and labor-consuming,this paper realizes the automatic prediction of the ear number,grain number and yield in 0.25m2 area based on the canopy image of wheat in the field.In this study,three wheat samples of two varieties at different mature stages were used in the field experiment,two samples were acquired from one of wheat variety with two different maturity stage.0.25m2 area of wheat field was gained by the sampling frame,and the wheat canopy image in field was obtained at a slope angle by camera above wheat canopy.In each sample area,five wheat ears were selected as markers to research the relationship between the ear area of a single wheat ear and the number of grains and between the area of a single wheat ear and yield.The image of a single wheat ear in vitro was obtained by using a scanner.The awns and stalk were removed by image processing.Pixel area of the ear was obtained after binarization,and the relationship between pixel area of the ear and the number of grains per ear and between pixel area of the ear and the weight of grains were analyzed respectively.The research results showed that there were linear relationships between the ear area and the number of grains per ear and between the ear area and the weight of grains with different varieties and different maturity stages,and the determination coefficients R2 were 0.8015 and 0.7823 on average.The image distortion correction and scale normalization of wheat canopy image in the field were carried out to extract the area of single wheat ear in the vivo and analyze the relationship between the area of single wheat ear and the number of grains per ear and between the area of single wheat ear and the weight of grains.The research results showed that there were the linear relationships between the ear area and the number of grains per ear and between the ear area and the weight of grains with different varieties and different mature periods.The determination coefficients R2 were 0.7371 and 0.7673 on average.Therefore,it was possible to predict the number of grains and weight of wheat by obtaining the ear area of wheat in the field.The image of wheat canopy was preprocessed by distortion correction and scale normalization..The color space of the image was transformed from to HSV,the sampling frame and background plate were removed according to the saturation component of HSV color space.The color space of the image was transformed from RGB to Lab,the contrast between the target and background in the image was enhanced by improved gamma transform processing the Lab image,the target and background were roughly segmented by K-means clustering algorithm,and the image was smoothed by median filter.The processed wheat canopy image was used as the prediction sample set of machine learning classifier,five images were selected respectively from each prediction sample set.The training sample set and its positive part of machine learning classifier was composed of the ear part of wheat which was extracted from image by PS software,and the classification label set was established.The machine learning classifier trained the wheat image in the training sample set and generated the prediction model.The prediction model generated by the classifier was used to process the wheat image in the prediction sample set and the ear part was extracted.The pixel area of the ear part was calculated after binarization,and the linear relationship between the pixel area of the ear and the grains and yield was established.The wheat ear was highlighted by contrast enhancement processing in the wheat canopy image removed sampling frame and background plate after preprocessing.The wheat awn and high frequency noise were removed by smooth filtering,and then the gray-scale processing was carried out.Finally,the local maximum points of gray-scale image were calculated by Find Max-ima algorithm in ImageJ,and the number of local maximum points was calculated and it was equal to the number of regional wheat ear.The results showed that the average absolute error of the number of regional wheat ear was 10 ears and the average relative error of the number of regional wheat ear was 9.72%.The average absolute error of the wheat prediction yield in 0.25m2 area was 12.37g and the average relative error of the wheat prediction yield in 0.25m2 area was 14.04%by using the predicted number of wheat ear,the average number of grains per ear and thousand-grain weight.The average absolute error of the predicted number of grains in 0.25m2 area was 209.32 grains,and the average relative error was 8.17%by taking the pixel area of wheat ear in 0.25m2 area into the linear relationship formula between the pixel area of wheat ear and the number of grains.The average absolute error of wheat prediction yield in 0.25m2 area was 5.3 0g,and the average relative error was 6.05%by taking the pixel area of wheat ear in 0.25m2 area into the linear relationship formula between the pixel area of wheat ear and yield.The average absolute error of wheat prediction yield in 0.25m2 area was 5.55g and the average relative error was 6.11%by using the predicted number of grains in 0.25 m2 area and thousald-grain weight.In general,accuracy of predicting regional wheat yield was highest when the relationship between ear area and yield was used.In this study,the number of ears,number of grains and yield of wheat in 0.25m2 area are automatically acquired based on the canopy image characteristics of wheat in the field and the prediction accuracy is above 90%.There are the advantages of strong operability,low economic cost and certain practicability in using this yield prediction technology. |
PDF Full Text Request