| It is significant that utilizing a Wireless Sensor Network(WSN)to continuously perceive crop growth information in real-time for regulating and controlling the growth of crops as well as producing high and stable yields.Crop growth information monitored by the sensors serves as the basis for decisions concerning the regulation and control of the growth of crops.Therefore,the accuracy of such information is vital.Considering that agricultural production features characteristics of vast areas,scattered fields,and long crop growth cycles,learning how to greatly reduce network deployment costs while guaranteeing full-coverage monitoring of crop growth information in the fields will directly affect the popularization and application of WSN in the agricultural industry.The National Engineering and Technology Center for Information Agriculture at Nanjing Agricultural University has developed a multi-spectral crop growth sensor called CGMD302.The sensor uses sunlight as its light source to collect canopy spectral reflectance of 720 nm and 810 nm.By integrating CGMD302 sensors,WSN can collect crop growth information,such as leaf nitrogen content,leaf nitrogen accumulation,leaf area index,leaf dry weight.Under field conditions,seasonal variations in temperature and sunlight cause the internal temperature of the sensors to vary across a large range.The solar elevation angle also varies constantly.Variations in temperature and solar elevation angle undoubtedly will have an effect on sensor reflectance.With regard to the aforementioned issues in the application of the crop WSN within the agricultural industry,this project has taken efforts to do research on compensation and deployment technology of the crop WSN.To advance the stability of the crop WSN nodes in farmland applications,temperature and solar elevation angle compensation models with low computation burdens and high precision were constructed.The results show that the node output voltage increases while the reflectance decreases with an increases in temperature.A temperature-based node output voltage prediction model has been established using the symbolic regression technique.By performing a form transformation on the model,the node output voltage could be corrected to the node output voltage at 25℃,and temperature compensation for the node could be realized by calculating the reflectance in accordance with the node output voltage at 25℃.The compensation enables reflectance fluctuation caused by temperature influence to be decreased from 1.0-7.0%to less than 0.45%.As the solar elevation angle increases,the reflectance of the node decreases and reaches the minimum value at 12:00 local time.The daily variation graph is U-shaped.However,the reflectance variation within the time period of 11:00 to 13:00 is too small to affect measurement precision.Optical energy received by the node’s upstream optical sensor was decomposed into direct light and scattered light from the sky.The De Wit approach and Berlage approach were used to estimate the respective ratio of the direct light and scattered light to the sky light at different solar elevation angles,and a standard sky model of International Commission on Illumination(CIE)was used to calculate the distribution of scattered light in the sky.Using such basis,in accordance with the optical path structure of the sensor,the direct light energy and scattered light energy transmitted into the sensor were calculated using the Fresnel formula.The output voltage of the upstream optical sensor was converted into irradiance of the node(equivalent to the irradiance of the crop canopy),and the output voltage of the downstream optical sensor was converted into reflected radiance of the measurement target,such as crop canopy.The reflectance could be calculated based on the reflective radiance and the irradiance,thereby realizing solar elevation angle compensation for the node.The solar elevation angle compensation model was established based on theoretical derivation,so it has good universality.After compensation,the variable coefficient for daily variation of reflectance decreases from 9-13.8%to 0.2%.Also,the daily variation presents a horizontal straight line,thereby efficiently reducing the influence of the solar elevation angle.Sensing nodes of WSN often use batteries as their power source.If the nodes need to frequently operate the temperature and solar elevation angle compensation models,energy consumption of the nodes will be accelerated,and thus shorten the network service life.With regard to this situation,the compensation model may be running on the gateway,and the upstream optical sensor of CGMD302 was integrated at the gateway,while the nodes of crop WSN were integrated the downstream optical sensors only.Due to the gateway’s strong operational capability and considering the convenience using of the compensation model,a BP neural network optimized by genetic algorithm was used to build a one-time temperature and solar elevation angle compensation model for the upstream optical sensor.Combined with temperature compensation model constructed above,the temperature and solar elevation angle compensation model of crop WSN reflectance was constructed.Variable-length chromosome coding and double crossover operation were designed for genetic algorithm(GA)to synchronously optimized the topology structure,weights and thresholds of the BP network(named as the GA-BP algorithm).Experimental tests show that relative error of the reflectance predicted by the model are mostly less than 0.6%,and mainly concentrated between 0%and 0.4%.Thus,the model has high compensation precision and good practicality.Full-coverage monitoring could obtain complete information of a target area and is the foundation for the application of the WSN.The deployment cost(number of nodes in the network)of the network is a factor that limits the popularization and application of the WSN.In the actual application,it is a contradiction between full-coverage monitoring and low deployment costs.In general,the more nodes were used,the more information could be collected,but also increasing the deployment cost.Differences in crop conditions are influenced by the spatial distribution of soil nutrients.If the nutrients are distributed evenly,the crop conditions are expected to be approximately uniform,with little difference throughout the farm;on the contrary,if nutrients are not distributed evenly,there will be great differences in crop conditions.In accordance with the differences in the spatial distribution of soil information in farmland,fuzzy c-means(FCM)clustering was applied to divide the farmland into several areas,where the soil fertility of each area is nearly uniform.Then,the crop growth information in the area could be monitored with complete coverage by deploying one sensor node to that area;hence,this has the potential to greatly decrease the number of deployed sensor nodes.Moreover,in order to determine the optimal cluster number of fuzzy c-means clustering,a discriminant function of Normalized Intra-Cluster Coefficient of Variation(NICCV)has been established.Multiple dataset tests with different spatial complexities indicate that NICCV can accurately determine the optimal cluster number.The sensitivity analysis indicates that NICCV is insensitive to the fuzzy weighting exponent,but shows a strong sensitivity to the number of clusters.The network deployment method based on soil spatial differences not only satisfies requirements for full-coverage monitoring,but also clearly reduces network deployment costs,thereby solving the contradiction between high coverage and low cost requirement in network deployment.The WSN is a multi-hop network.Some of the nodes(relay nodes)function to forward information of the other nodes.The failure of a few relay nodes may cause an interruption with communications of the other nodes.Therefore,in order to advance the robustness of network communications,the nodes in the network should have at least k connections.In addition,the information sensed by the nodes will be meaningful only when the nodes are located within the target monitoring area,and thus the network’s information transmission will be necessary.Thus,the network coverage should be considered first during the network’s connectivity deployment.Farmland is characterized by irregular and scattered fields as well as vast areas,and a digital map could clearly show the farmland’s spatial information characteristics.This project realizes the deployment of the WSN with high robustness information transmission across vast farmland by using Genetic Algorithm(GA)according to the spatial information provided by the farmland digital map.The adaptability of GA was calculated based on the following network deployment principles:(1)The nodes should be deployed within the corresponding plots,and always keeping a certain minimum distance from the boundaries to avoid the influence of boundary effect;(2)the nodes should be kept in κ-connections,in order to advance the robustness of network communications;(3)there should be no communication silos.To improve the algorithm’s performance,a buffer area operation of GIS was performed inwardly through each plot,in order to obtain the deployment area of the nodes,and the principle(1)would be realized by developing the nodes within the area.The connectivity number of a node could be obtained by calculating the number of neighboring nodes for each node.During network deployment,communication silos can be avoided by pursing large connection angles.By using analytic hierarchy process,the described multi-target optimization has been transformed into a single-target optimization issue.This has provided and established an adaptability function for the genetic algorithm and developed a wireless sensor network deployment system.Experimental tests showed that the crop WSN deployment system boasts excellent performance for network deployment of both regular and irregular farmlands and completely satisfies requirements for agricultural production.Experiments for different transmission distances between sensing nodes across a total of 90 plots of common farmland in Jiangning,Nanjing showed that the proposed algorithm is highly efficient and can be successfully used for wireless sensor network deployment. |
PDF Full Text Request