图像分割方法应用于施工现场物体的识别

Application of Image Segmentation Method Based on Depth Learning on Object Recognition in Construction Site

复杂场景中的图像分割是当前图像分割中的一个难点,给分割算法带来了更大的挑战。基于深度学习的算法基于统计学理论,相比传统的神经网络,深度学习能够进行更深层次的学习,因此准确率大大提升,本文研究了一种深度信念网模型,加入drop out策略,并且进行改进,最后把模型应用于施工现场勾机的图像分割与识别。实验证明,改进的深度信念网模型算法可以有效的识别复杂场景中的图像。

Image segmentation of complex scenes is a difficult problem in the current image segmentation subject, which brings more challenges to the segmentation algorithm. The deep learning algorithm based on statistical learning theory, compared to the traditional neural network, can learn the deeper level content of image, so the segmentation accuracy is greatly improved. This paper studies the deep belief network model which is one deep learning model, adding the drop out strategy to make improvements, finally, the model is applied to image segmentation and identification of hook on the construction site. The experiments show that the improved depth belief network model can effectively segment and identify the object in the complex scenes.