| In the process of crop growth,dynamic monitoring of crop phenotypic parameters can timely and accurately understand crop growth information and nutritional status,which is convenient for later management of water and fertilizer,so as to ensure the normal growth of crops and meet the strategic goal of sustainable development.Leaf moisture content can fully reflect soil moisture,environmental conditions,and crop moisture status,and can further influence crop yield and quality by guiding irrigation.Therefore,leaf moisture content can be used as an important indicator for crop phenotypic monitoring.In the process of growth,it is of guiding significance to control the nitrogen status of plants,especially leaves,and improve the yield and quality of fertilizer and water.Therefore,this paper took the canopy image of wheat as the research object,and applied the artificial neural network technology and image processing technology to study the leaf water content model and nitrogen monitoring.The main work and innovations are as follows:1)Research on adaptive weighted particle swarm optimization algorithm to improve the traditional K-means image segmentation method.The strong global search ability of adaptive weighted PSO algorithm was used to determine the initial clustering center of the image,and then the image was segmented according to the strong local search ability of K-means clustering algorithm.The segmentation effect,stability and environmental adaptability of the algorithm were verified by images with different pixels and different weather conditions,which were better than the traditional K-means clustering algorithm and Otsu algorithm.On the basis of image segmentation,crop coverage data are calculated.The total relative error and root mean square error of the algorithm proposed in t HSI paper are both lower than 0.2,significantly lower than the traditional K-means algorithm.2)Research and design of non-destructive monitoring model of wheat leaf water content based on BP neural network.In the natural environment of field,a variety of sensors were used to collect wheat leaf images,and color,texture and temperature features of different images were extracted respectively.A total of 13 feature parameters were extracted,including 6 color indexes,6 texture indexes and canopy temperature.The correlation between the above parameters and the measured water content was analyzed,and the fusion of 7 characteristic parameters with high correlation was selected as the input information of BP model.The prediction model of water content of leaves was established by training the BP neural network.After performance verification,the accuracy of the leaf water content prediction model reached 98%,and the correlation between the predicted value of the model output and the true value was as high as 0.994.The leaf water content prediction model is integrated with multi-source image information,which can accurately,quickly and non-destructively predict the water content of wheat leaves.3)The diagnosis of nitrogen nutrition in wheat was studied by multispectral imaging.The differences in nitrogen nutrition in wheat were manifested as the differences in chlorophyll content and biomass of the crops.The study found that there was a very significant positive correlation between the nitrogen application amount of wheat and chlorophyll content and biomass,and the correlation coefficient was 0.820 and 0.751,respectively.According to the commonly used vegetation index in the calculation of spectral reflectance of different bands of wheat canopy,the correlation analysis was carried out with the conventional nutritional diagnostic index,and the linear regression model of wheat nitrogen was established.In actual production,the chlorophyll content and biomass of wheat could be estimated indirectly by the vegetation index to diagnose the nitrogen nutritional status of wheat. |
PDF Full Text Request