| Since the 1990s,monitoring crop growth information has become an important technology for smart agriculture.Accurate and rapid acquisition of crop growth status can provide not only reliable information for agricultural management but also prior knowledge and specific parameters for crop yield estimation.Agricultural Internet of Things employs various communication devices and sensors to collect crop growth data.The data are transmitted to the crop growth monitoring platform to realize the storage.Consequently,the real-time control of agricultural factors can be implemented.Furthermore,crop image is one of the important carriers of crop information,which directly indicates the crop growth status.Thus image is also the important data source to crop growth monitoring platform.A remote crop growth monitoring platform was developed to realize real-time remote monitor of growth data including images and environmental parameters.The platform can collect traditional environmental data and crop images.The requirement of storage space is high while storing the original crop images.Therefore,the compression method of crop image was investigated in this paper.According to the characteristics of crop images,a compression algorithm of region-of-interest(ROI)was proposed.Firstly,the segmentation method of vegetation image was studied.A low contrast image segmentation method based on color factor and superpixel clustering was proposed.After ensuring accurate crop segmentation area,the vegetation area compression method based on superpixel and clustering was ultilized to achieve the high efficiency and high quality compression requirements.The main research contents are as follows:(1)In order to monitor the crop growth and control the agricultural machinery in greenhouse,a remote platform is developed.The platform adopts a three-layer Io T structure,including the perception layer,network layer and application layer.Sensors in the perception layer collect the environmental parameters and the crop images.The data are transmitted to the network layer according to the specified communication protocol.By employing the Vue+Spring Boot framework,the business logic processing,data storage analysis and page interaction are implemented in the network layer.Through the parallel communication by TCP and UDP,the interaction and data reception do not interfere with each other.Thus the service coupling degree of the system is reduced.In the application layer,five modules are designed including data center,control center,equipment center,video center and setting center.Through functional tests,it is verified that the platform can meet the demands of remote monitoring the crop growth and controlling the agricultural machinery in greenhouse.(2)Vegetation image segmentation is of great significance for smart agriculture.K-means algorithm,as an unsupervised learning method,has been widely applied in the vegetation image segmentation.But it is difficult to achieve accurate segmentation of vegetation images under low contrast conditions.Noise points bring the errors to the segmentation.Several image filters and morphological operators are employed to remove the noise points.However,over-or under-segmentation often occurs.To address the issues,a vegetation image segmentation algorithm based on color index and superpixel clustering was proposed.Firstly,the color index was utilized to enhance the color feature of crops.Secondly,the superpixels were generated by linking the adjacent pixels with the similar luminance feature based on SLIC algorithm.The procedure was capable of removing the noise points and preserving the crop areas effectively.Finally,the superpixels were classified into vegetation regions and background regions based on K-means algorithm.Thus the segmentation was obtained.In the experiments,97 low-contrast images of sugar beet were utilized to evaluate the proposed method by comparing with the common color indices in subjective and objective aspects.The average PRC,RC,F1,ACC and SPC of our method achieved 94.78%,93.42%,94.04%,99.55% and 99.80% respectively.The scores were higher than the other methods except that the PRC and SPC were lower to that of Ex GR.Nevertheless,the subjective evaluation result of our method was better than that of Ex GR in terms of overall segmentation and regional details.The results proved that the proposed method could achieve an accurate segmentation of crop images in low-contrast scenes.(3)Finally,after the location information of vegetation and background area is obtained by image segmentation,the compression method of ROI was explored.DCT method,superpixel algorithm SLIC and clustering algorithm K-means were employed to compress vegetation images.By comparing the experimental results under different parameters,the compression quality and efficiency of vegetation image compression were analyzed.A vegetation ROI compression algorithm based on superpixel and clustering method was proposed.Firstly,the vegetation and background regions were located.K-means was ultilized to compress the vegetation region and retain the color,shape and texture information.Then the background region was compressed with high compression ratio by SLIC.Finally,the compressed vegetation and background regions were merged.The experimental results proved that the proposed method can meet the requirements of image quality and compression efficiency. |
PDF Full Text Request