Design And System Implementation Of Penguin Habitat And Quantity Detection Method

Under the impact of global warming,the tendency of “greening” has been threatening Antarctic Peninsula.As the largest species in biomass on Antarctic,penguins are known as the bio-indicator of the climate change in Antarctic.Researches in penguins' population and the distribution of their habitats provide basic data and theoretical guidance to studies in climate change's impact on penguins and the species' preservation.Therefore,accurate and efficient monitoring on penguins' habitats and population is of vital importance to the study of penguins.As deep learning technology develops,monitoring penguins' habitats and population with this new technology has become an important approach of acquiring relevant information.The researcher determined the geographic distribution of penguins' habitats through object detection technology and conducted test statistics on penguins' population in habitats through semantic segmentation,in turn,results of penguins' habitats and its population was obtained by quantity surveying,offering statistic backup for preservation of penguins and Antarctic's environment.Steps to fulfill the task are as follows.(1)Preprocessing of data drew from graphics of penguins' island habitats.Preprocessing involves image segmentation,bulk image naming,expanding and sorting sample quantity and so forth.Preprocessed images,after data annotation,became data set needed for algorithmic model of this research.(2)The object detection of this research was based on Faster RCNN.Considering the missed and false detection during the process and the difficulty in recognizing the results,the researcher designed and realized a detecting method on the basis of Faster RCNN,which replaces the region of interest pooling in Faster RCNN with ROI Align,increasing the accuracy of detecting.Besides,this method makes users recognize results more easily by replacing detection box labelling with region image labelling.(3)The detecting method based on Fully Convolutional Networks(FCN).Conducting the detection by inputting the processed detecting results of penguins' habitats,combining with FCN technology,the efficiency of extracting individual penguins from original images was improved.(4)The design and realization of penguins' habitats and population detecting system.Based on methods mentioned above,the researcher built a system to detect penguins' habitats and population.The system can produce detecting results of penguin habitats and population by inputting images needed for detecting.After testing modules of the system,the results showed all modules have reached desirable effectiveness.Therefore,it can provide data on penguin habitats and population for users with high efficiency and accuracy.