Drought Detection Of Maize Plants Based On Machine Learning And Computer Vision

In recent years,with the rapid development of computer vision technology,more and more image-based applications have been developed to solve the problems in agricultural production.At present,plant drought stress is the most important factor affecting crop growth and yield.Therefore,maize drought detection is conducive to arranging reasonable irrigation in time and ensuring final yield when maize is under drought stress.At present,the detection of maize drought mainly depends on artificial observation in most agricultural production areas.With the development of agricultural automation,many field data can be detected in real time,such as leaf temperature,soil moisture,chlorophyll content,etc.These data can be used to assist in understanding the degree of maize drought.However,to obtain these real-time data,a large number of sensors need to be laid in the field and have a high cost of the instruments.Based on these background,this paper takes potted and field maize as the main research objects.and from the perspective of computer vision to provide an effective solution for drought detection.The work of this paper can be summarized as follows.(1)Automatic measurement of the number of plant leaves in the laboratory potted environment.The traditional convolutional neural network uses the direct regression algorithm to count maize leaf number,which has a large error in the actual experiment.Therefore,this paper introduces the method of CNN and Fisher vector coding to extract multi-scale features.In the design of feature extraction network,we refer to the multiscale convolution core structure of Google Le Net to make it more suitable for extracting the characteristics of blades of different sizes.Then,Fisher vectors are used to encode the middle-level feature maps to enhance the ability of feature expression.Finally,the number of leaves was regressed by random forest,and the effects of different drought degrees on the number of maize leaves were demonstrated in the experiment.(2)Drought detection in mid-growth stage of Maize based on visual features was researched.In order to simulate the decision-making mode of agricultural experts,the single maize plant was described from three aspects of texture,color and morphology,and the features of average leaf inclination and leaf dispersion were first time proposed to reduce the effect of illumination on image texture and color.The test results show that the model with morphological features has a higher recognition accuracy.Subsequently,based on the above three features and the change rate of leaf number with the growth cycle,LSTM was used to predict the periodic drought of maize plant.The final experimental results show that this method can effectively predict the drought degree of maize plant in a certain periodic sequence.(3)Drought detection in mid-growth period of maize population under complex field background was considered.Field environment is more complex than laboratory environment,such as natural light,attitude disturbed by wind direction and wind speed,leaf occlusion and overlap,so it is impossible to use single plant detection method.In this paper,the idea of global feature extraction is proposed,and the texture and morphological features of multi-plants are preliminarily extracted by using multi-angle and multi-wavelength Gabor filter.Then a lightweight CNN with fewer training parameters is designed and the powerful ability of extracting local features by convolution calculation is used for feature integration and classification recognition.Finally,a high recognition rate is achieved on the test set.In summary,the maize drought detection method proposed in this paper has good adaptability to complex environment,and solves the problem of drought automatic detection of single and group maize plants.The research results of this paper have important application prospects for agricultural production automation.