Design And Implementation Of Intelligent Measurement System For Grassland Coverage Based On Mobile Terminal

Grassland coverage is one of the important evaluation indicators of grassland ecosystem health.Accurate measurement of grassland coverage is of great significance to the protection,management and utilization of grassland ecosystems.Aiming at the measurement efficiency of grassland coverage field survey in ecological monitoring,this thesis designs a set of intelligent grassland coverage measurement system based on the commonly used mobile terminal Android framework.In order to effectively overcome the interference of shadows,fallen leaves and other ground objects on grassland image recognition,the system adopts a deep learning method that combines multi-dimensional feature information of grassland images.In addition,in order to solve the problems of accurate measurement,data management and easy operation,the system combines PC-side server and mobile terminal equipment.The system can serve for ecological field investigation and monitoring.The research content of this thesis mainly includes:(1)Construction of grassland image label dataset.Aiming at the problem that grassland images are difficult to label manually,this thesis collects four types of image data: broad-leaved grassland,coniferous grassland,deciduous coniferous mixed grassland and moss-covered vegetable field.Using a semi-automatic labeling method that combines threshold segmentation and image translucent superposition,combined with data enhancement methods,it effectively solves the problem of small sample data expansion,and improves the accuracy and production efficiency of data labels.(2)Research on grassland image recognition model.Aiming at the problem of low segmentation accuracy due to the complex background of the grassland image,the variable size of the plant target,and the different shapes of the grassland leaves,this thesis introduces two attention mechanism modules and constructs the R-Unet network based on the idea of an encoder-decoder.The network model uses an improved pyramid pooling module and multi-scale fusion at the end of the backbone feature extraction network to enhance its feature extraction capabilities.In the decoding stage,the skip connection is used to splicing and fusing the encoding pooling results of the backbone network with the upsampling features in the decoder to improve the ability to capture the detailed contours of leaves.The R-Unet model built in this thesis has an MIou of92.41% on the grassland image dataset.Compared with the mainstream basic model with the best effect,its accuracy has increased by 5.16%,and the recognition and segmentation effect has been significantly improved.(3)Design and implementation of a grassland image coverage measurement system based on mobile terminals.The thesis divides the entire system design into front and back ends,the front end is used to display static resource pages and recognition results,and the functions of the back end are image processing and data management.The system design focused on solving the problems of user identity authentication query,system high-availability structure optimization and system queue flood discharge.Through the field test of multiple functional modules of the system,and the performance pressure test of the core interface of the system through simulated traffic,the results show that the availability of the system has achieved the expected effect.