Research On Compression And Transmission Method Of Wildlife Monitoring Image Based On Target Region Extraction

Strengthening wildlife monitoring is conducive to preventing poaching and comprehensively implementing laws and regulations prohibiting the consumption of wildlife.As one of the main methods of wildlife monitoring,wireless multimedia sensor networks can obtain the image information of the monitoring wildlife,and can accurately estimate the species diversity and population number,and realize long-term real-time monitoring,thus providing a scientific basis for the protection of wildlife resources.The contradiction between the large amount of data of monitoring image and the limited power consumption of the monitoring carrier sensor network is increasingly prominent.In order to more effectively compress the wildlife images and meet the image quality requirements of data analysis,this paper proposed a novel compression and transmission method for wildlife monitoring image,and makes in-depth research from three aspects: target region extraction,compression and transmission strategy research and compressed image restoration.The main contribution is as follows.1.A method of target region extraction for wildlife monitoring images was proposed in this paper.Based on the adaptive mean-shift algorithm,combined with color space reconstruction and Hermite filter construction,a method of target region extraction suitable for wildlife monitoring images was proposed.This method achieved the optimization of the wildlife monitoring image foreground areas extraction and improved the accuracy of target region extraction.2.The compression and transmission strategy of wildlife monitoring image was optimized.Based on the extraction of the target region of the wildlife monitoring image,combined with the regionalcharacteristics of the binary mask,the mask coding method was improved,and the coding code stream was saved;the paper further analyzed the bit-planes transmission method and the set partitioning in hierarchical trees(SPIHT).By setting unimportant bit-planes and adding set judgments,on the basis of ensuring the image reconstruction quality of the target region,the proposed method achieved priority transmission of the target region and reduced data redundancy.3.The restoration algorithm of wildlife compressed images was improved.In order to improve the performance of the network,this paper used the generated adversarial network as a framework,embedded the improved Squeeze-and-Excitation block into the generator and discriminator,and used VGG19 as a feature extractor,combined with the mean square error(MSE)and adversarial loss,to optimize the loss function.The proposed method realized the restoration of the wildlife monitoring image from compressed image to high-resolution image,improved the peak signal-to-noise ratio(PSNR),and restored high-frequency and texture parameters.