Research On High-performance Image Semantic Segmentation Algorithm In Complex Environments

Image semantic segmentation is one of the most fundamental techniques for visual intelligence,which plays a vital role for mobile robot tasks such as scene understanding,autonomous navigation,and dexterous manipulation.However,semantic segmentation of indoor environments poses great challenges for existing segmentation techniques due to the complex overlaps,heavy occlusions,and cluttered scenes with objects of different shapes and scales,which may lead to the loss of edge information and insufficient segmentation accuracy.And most of the semantic segmentation networks are very complex and cannot be applied to mobile platforms.Thus,it is of significant importance for ensuring as few network parameters as possible while improving the detection of meaningful edges in complex scenes.In this paper,we propose a new semantic segmentation method for mobile platforms working in complex environments.Our approach systematically incorporates convolution features extracted from each layer of the convolutional network,so that the network can make inferences under the rich high-resolution feature maps.The main contributions of this article are summarized as follows.(1)We propose a new module for acquiring the hidden convolutional feature,which can extract useful detailed feature information from all the hidden convolutional layers of CNN.(2)In order to enrich the convolution feature map and process the edges of the target objects more accurately,we use dilated convolutions with different sampling rate to form a pyramidal convolution module,which extracts multi-scale high-level semantic information of images and integrates with the fine-grained features extracted by the hidden convolution layer.(3)We present an innovative image semantic segmentation method architecture that achieves a balance between accuracy and network complexity,and proposes a new solution for image semantic segmentation in complex environments.Our method refines the segmentation of object boundaries in complex environments while ensuring as few network parameters as possible to achieve a balance between model accuracy and complexity,which is a high-performance end-to-end multi-scale convolutional features network.Finally,the resulting approach is extensively evaluated on the prestigious indoor image datasets of SUN RGB-D and NYUDv2,and shows a good performance on the semantic segmentation task of complex scenes.Our method can achieve the improvement of accuracy while guaranteeing fewer parameters.Compared with the latest representative models Deep Lab v3+ and Hr Net V2,our method shows certain advantages in accuracy and parameters.