| Nowadays, traditional agriculture is rapidly transforming to modern agriculture in our country. Agricultural production will become more intelligent with the deep integration of information technology and agriculture, which is a trend of modern agriculture development. Under the background of the development of modern agriculture, the evaluation of wheat growth state based on image analysis technology were studied in this paper, and a series of new methods to realize the intelligent and efficient management of wheat production were proved. In the main line of the wheat growth process, we investigated the calculation method of the main agronomic parameters in wheat seedling stage, wintering stage, jointing stage, booting stage and mature stage, the evaluation model of wheats growth state was established, and the main agronomic parameters of wheats were acquired intelligently. Finally, we completed the group growth state intelligent evaluation system. The results can provide technical support and theoretical basis for the intelligent monitoring system in the Internet of things and provide reference for the development of intelligent field measurement and evaluation software based on mobile terminal. The main conclusions are as follows:(1) In the seedling stage, automatic wheat seedlings counting method was developed. Wheat field seedling density has a significant impact on the yield and quality of grains. Accurate and timely estimates of wheat field seedling density can guide cultivation to ensure high yield. The objective of this study was to develop an image-processing based, automatic counting method for wheat field seedlings, to investigate the principle of automatic counting of wheat emergence in the field, and to validate the newly developed method in various conditions. Digital images of the wheat fields at seedling stages with five cultivars and five seedling densities were acquired directly from above the fields. The wheat seedlings information was extracted from the background using excessive green and Otsu’s method. By analyzing the characteristic parameters of the overlapping regions (Overlapping region is a number of overlapping wheat seedlings in the image) of the fields, a chain code-based skeleton optimization method and corresponding equation were established for automatic counting of wheat seedlings in the overlapping regions. Fifty images representing five wheat cultivars with five seedling densities were counted using the newly-developed method. The results showed that this method could effectively count the seedlings in the wheat field with an average accuracy of 89.94%. Among the five seedling densities tested, the counting accuracy was highest (97.14%) for the wheat field with 135×104 ha-1 seedling density. Among all five cultivars tested, the counting accuracy was highest (92.54%) for Yangnuol. The seedling density significantly affected the counting accuracy, while the different cultivars had little effect. The results showed that the newly developed method can effectively count the number of wheat seedlings, with an average accuracy rate of 89.94% and a highest accuracy rate of 99.21%. The results also indicated that the accuracy of counting was not affected by different cultivars. However, the seedling density had significant impact on the counting accuracy (p<0.05). When the seedling density was between 120×104 ha-1 and 240×104 ha-1, high counting accuracy (>92%) could be obtained. The study demonstrated that the newly developed method is reliable for automatic wheat seedlings counting, and provides a theoretical perspective for automatic seedling counting in the wheat field. The method reported here could also be adapted for rice and other gramineous plant counting.(2) The estimation model for agronomic parameters was constructed in winter stage, jointing stage and booting stage. Non-destructive acquisition agronomic parameters of wheat growth status and appropriate evaluation are important to wheat management. This study was performed to construct a model for the estimation of wheat dry weight (DW). leaf area index (LAI), tiller number (TN), and nitrogen accumulation (NC) using image analysis techniques. The wheat cultivar used here was Yang waxy wheat 1. Two-factor randomized block experiment was designed. The study included five planting density levels and four levels of nitrogen fertilizer treatments. The ultra-green values (ExG) and Otsu method were used to remove the background of the wheat field image. The ratio of the number of pixels of the wheat to the total number of pixels in the image was used to represent wheat canopy cover (CC). Eight methods of calculating image features were used to assess the image features of wheat, which can be used to analyze the wheat growth status at different stages of growth. Prediction model for agronomic parameters were then established using stepwise multiple regression (SMLR) method, and the performance of the model was evaluated using coefficients of determination of models (R2), the root mean square error in prediction (RMSEP), and the relative error in prediction (REP). R2 and RMSEP were used to evaluate the stability of the model and the average deviation between the measured values and the true values. REP was used to measure the accuracy of the model with respect to prediction. The estimation model for agronomic parameters was then constructed using the stepwise linear regression method. The results show that estimation model proposed in this study offers more accurate simulation of agronomy parameters than the single-parameter model; the R2 values of DW, LAI, and NC are all above 0.8, and the TN’s reached 0.72. The four established models performed well with respect to predicting DW, LAI, TN, and NC and showed high R2 values and low RMSEP and REP. For the prediction of the four agronomic parameters in the modeling data set, the R2 values were between 0.77 and 0.91, REP was 15.46%-22.53%; for the validation data set, R2 was 0.72-0.85, and the REP value was 17.31%-21.26%.(3) In the Mature stage, automatic wheatears counting method was developed. Number of wheatears m-2 is the main component of grain production estimate. In order to calculate the number of wheatears in certain parts intelligently, an In-field wheatear counting method based on image analysis technique was designed. Firstly, five color features such as normalized difference index were analyzed to get suitable features which were used to extract wheatear from original image. Secondly, a comparison of the five texture features (Energy, Contrast, Homogeneity, Entropy, Relation) was performed and appropriate features were selected to segment wheat images. Finally, calculate the number of ears, in this step, erosion and dilation operations in binary mathematical morphology were performed so as to clear impurities and awns. Hole-filling algorithm and thinning algorithm were used to get unbroken wheatear and its skeleton; corner detection algorithm was selected to get the corners of skeleton with the purpose of estimating the wheatear number of connected region. The advantages and disadvantages of color segmentation and texture segmentation are analyzed deeply (include what environment do they well for and how long do they need).20 images with 71×92 pixels were used to evaluate the run-time of color segmentation and texture segmentation, the former take 16.97 milliseconds and the latter take 17.76 seconds. To validate the effectiveness of this method,20 drilling wheat images and 20 broadcasting wheat images were tested. And drilling wheat average counting accuracy rate is 95.63%, broadcasting wheat is 97.07%. The experimental results show that color feature and texture feature can be used to extract wheatear from original wheat image, color segmentation is faster than texture segmentation but less environmental adaptability; the corners of skeleton have close relationship with the number of wheatears in connected region.(4) The evaluation model for wheat group growth status was built using BP neural network. The evaluation of wheat population growth state can reflect the status of the wheat groups and provide new evidence for wheat non-destructive diagnosis and field management. In addition, the wheat population growth state evaluation model is composed of two parts of study:1) The agronomic parameters features at different growth stages of wheat population with different yield should be ascertained, and the agronomic characteristics of high-yield groups should be studied; 2) A BP neural network model should be constructed to determine wheat group growth status.In the first part, the wheat cultivar used here was Yang waxy wheat 1, and two-factor randomized block experiment was designed. The study included five planting density levels and four levels of nitrogen fertilizer treatments. Main results for this study are as follows:1) Study the change trend of yield with different seeding density; 2) Discern the influence of dry weight(DW), leaf area index(LAI), tiller number(TN), and nitrogen accumulation(NC) on wheat yield at winter stage, jointing stage and booting stage; 3) Study the change trend of yield with different wheatear numbers.The wheat population growth state evaluation model was established using these agronomic parameters. According to previous studies, wheat group growth status at wintering, jointing, and booting stages are closely related to DW, LAI, TN, and NC. In this study, the sample groups were divided into 13 grades through four K-means clustering. The evaluation model for wheat group growth status was built using BP neural network. The basis of evaluation in each stage are as follows:1) Seedling number was used to evaluate the reasonableness of planting density at seedling stage; 2) The wheat dry weight, leaf area index, tiller number, and nitrogen accumulation were used to evaluate the population growth state at wintering stage, jointing stage and booting stage.3) Wheatear number was used to evaluate the reasonableness of yield components at mature stage. In this study, the nonlinear mapping relationship between the various indexes and the growth status and the determination of the contribution rate of various indexes is the difficulty of evaluating wheat group growth status, which was solved. The average R2 value of the analog values and the values measured using the wheat population growth state evaluation model based on BP neural network were 0.83. The wheat population growth state evaluation model established using these four agronomic parameters can reflect the status of the wheat groups and provide new evidence for wheat non-destructive diagnosis and field management.(5) The software system of agronomic parameters measurement and growth state evaluation was developed. The software system is the specific practice of the previous algorithm; it is a useful way to evaluate the wheat population growth state. The software system was designed based on a three-layer C/S. We use Microsoft Visual studio 2013 as the development environment, and image processing toolboxes and computer vision toolboxes of MATLAB2014a were used to process images, and SQL Server2013 was used to establish databases. The software system could be used to count wheat seedlings and wheatears, it could also be used to estimate dry weight, leaf area index, tiller number, and nitrogen accumulation at winter stage, jointing stage, and booting stage. What’s more, the software system could evaluate wheat population growth state. This software system may offer some references for wheat growth intellisense and cultivational decision. |
PDF Full Text Request