A Semantic Segmentation Algorithm For Studying The Distribution Of Water Grass In Landscape

Landscape water grass planting density has crucial effect on water quality of repair,the proper planting can be more rapid repair contaminated water,artificial pruning growing grass time-consuming too loose or too tight,convenient sampling instrument automation processing water plants,so need to design a kind of can identify the semantic segmentation algorithm of hydrophytes distribution to guide automation machinery operation.Traditional image processing methods have single operation and poor robustness.In this paper,a semantic segmentation algorithm based on ASPP structure is proposed.This algorithm not only has significant segmentation effect on water grass dataset,but also performs well in the public dataset,surpassing the traditional algorithm and deep learning algorithm.Algorithm in this paper,first of all,on the basis of the FC-Dense Net network was improved,the original network feature extractor was added to improve the segmentation effect,add the network of feature extraction and the different structure of the FC-Dense Net contrast experiment on the Cam Vid dataset,the accuracy of amplitude,so add feature extractor has a promoting effect for improving the model segmentation effect.So in adding features,further improvement on the extractor of model using ASPP structure instead of the traditional convolution kernels,the improved model is firstly tested on the self-built water grass dataset,based on the evaluation index of the four different and the segmentation effect compare with the result of the experiment,the results show that FC-Dense ASPP has more superiority than other methods,can accurate segmentation does not contain sediment area of the grass.On the self-built datasets,the classification of test sets achieved 76% accuracy.In order to prove the generalization performance of this model,it was verified with other models on the Cam Vid open scene detection dataset and achieved 92% segmentation accuracy.