Research On Image Semantic Segmentation Algorithm In Lawn Scene

With the urban development,the area of green lawns in cities continues to increase,and the work of lawn maintenance is increasing.Therefore intelligent lawn mower is required.Image semantic segmentation is a pixel-level segmentation of various types of objects in the image scene.It is one of the basic and key techniques in the research of intelligent lawn mower.It can accurately identify obstacles in the environment and identify areas that need to mow.However,the traditional image semantic segmentation algorithm has the disadvantages of large volume and slow processing speed.In response to these problems,image semantic segmentation in lawn scene algorithms based on convolutional neural networks had been studied deeply.An image semantic segmentation algorithm based on lightweight neural networks and an improved image semantic segmentation algorithm based on the dual attention model were proposed.The main research and results of this paper are summarized as follows:(1)Segmentation dataset of Lawn scene.Firstly,aiming at the problem of the lack of public lawn scene segmentation datasets,lawn scene images were obtained through field framing and network acquisition of pictures.Then the data augmentation method was used to strengthen the acquired data and expand the number of pictures in the dataset.Finally,with the commonly-used image semantic segmentation datasets,the dataset images were manually annotated to establish a relatively complete dataset for image semantic segmentation in lawn scenes.(2)Research on Image Semantic Segmentation Algorithm Based on Lightweight Neural Network.Aiming at the problems of current image semantic segmentation algorithms with large volume and slow calculation,this paper proposed an image semantic segmentation algorithm for lawn scenes based on lightweight neural networks.Depthwise separable convolution was used instead of the standard convolution method,and the network parameter amount was greatly reduced on the premise of ensuring a certain segmentation accuracy.Then,a small size convolution kernel was used to further reduce the network volume.Experimental results show that the algorithm in this paper can achieve image segmentation speed with fewer parameters while maintaining good segmentation results.(3)Research on Image Semantic Segmentation Algorithm Based on Attention Model.In view of the above-mentioned accuracy degradation caused by the lightweight model,this paper proposed an image semantic segmentation algorithm combining a dual attention model.Combining the spatial attention model with the channel attention model and using a parallel approach to build a dual attention model,this model can be transplanted into other algorithm models as a module.Experimental results show that the algorithm in this paper can improve the segmentation accuracy of the model while maintaining the speed of segmentation,and also prove the effectiveness,real-time and practicality of the algorithm in this paper.