Design And Implementation Of Vision Based Intelligent Agriculture Management Platform

With the continuous development of information technology and Io T technology,China’s agriculture has entered an era of intelligence.Io T devices can be used to collect data and form large data centers.However,the main problem currently is how to use this data to provide decision-making basis for agricultural production management and development.Therefore,this article aims to use computer vision technology to mine and analyze agricultural data and to study the following aspects based on the smart agricultural management platform as the research object:(1)In response to the clustering phenomenon caused by missing measurement values in the fertilization decision-making model in soil moisture monitoring,this study proposes a new two-stage missing feature density peak clustering algorithm called DPC-INCOM.The algorithm first uses the density peak clustering algorithm on data with complete features and trains a classifier using labeled core points to represent the entire data distribution.Then,it calculates the symmetric FWPD distance matrix for incomplete data points and rescales and classifies them.The experimental results show that the error between the fertilization formula analysis using the DPC-INCOM algorithm and the actual fertilization is less than 0.25%.This method has good performance in crop moisture monitoring.(2)In response to the problem of complex pest identification methods and relatively low classification accuracy in existing pest monitoring,this study proposes a brand-new end-to-end Deep Pest Net model for pest classification and identification.Through experiments on collected crop pest images,the Deep Pest Net model has been found to have good performance.The overall accuracy,precision,recall,and F1＿score are 97.98%,98.48%,98.21%,and 98.33%,respectively,indicating that the model performs well in pest classification and identification and is significantly better than the latest CTCSP,Trans FG,NSC,and other methods.(3)In response to the low efficiency,long time-consuming,and high management costs of traditional manual visual disease diagnosis methods,this study proposes a deep learning-based multi-class crop disease target detection model.By conducting experiments on four rice diseases(rice blast,rice blast,bacterial blight,and brown spot),it was found that the Io U of the model in this study was 13.5% higher than that of the original YOLOv4 model.The model has been optimized in terms of detection speed and accuracy,and has good application prospects in the field of agricultural disease diagnosis.(4)In response to the difficulties of different crop growth in different regions,long growth cycles,and the time and cost of field experiments and crop data collection,this study proposes a new solution,which is to use a time convolutional network domain-adversarial neural network architecture to accurately predict plant growth curves with less training data in the target domain.Through experiments on collected corn growth data,it was found that the method based on TCN was superior to that based on LSTM,and the minimum value of RMSE was 2.621,indicating that the method performed well in predicting the accuracy of time series data.(5)This study has built a smart agricultural management platform,which includes the following subsystems: large field monitoring subsystem,video monitoring and linkage subsystem,agricultural product traceability subsystem,and expert guidance subsystem.The platform is programmed using Java language,and the basic and intelligent management implementation effects of the platform are demonstrated by showcasing the running effects of typical functions.