Research On Feature Extraction And Classification Algorithm For Point Cloud Recognition Of Safe Location Of Dangerous Chemicals Warehouse Goods

With the rapid development of industrialization,the demand for raw materials in the chemical industry in China has also rapidly increased.Among them,the safety position monitoring of goods stacking in hazardous chemical warehouses is an urgent problem to be solved.3D point cloud safety position detection based on depth cameras is a promising method.However,the characteristics of warehouse space being open,point cloud distribution being sparse,and a large number of noise points have led to feature extraction of point clouds Classification and model migration face many problems.The main work of this article is as follows:(1)A data preprocessing method based on point cloud denoising and adaptive down-sampling is proposed to address the problem of large amounts of point cloud data for hazardous chemical goods,as well as multiple and repetitive noise points.Firstly,the Ran SAC method and radius filtering are used to remove most of the noise from the point cloud data.Then,an adaptive down-sampling algorithm based on point density estimation is used to simplify the data and remove some redundant points,The experimental results indicate that the data preprocessing method proposed in this paper can effectively improve the classification accuracy of the point cloud classification model.(2)A point cloud feature extraction and classification algorithm using spatial feature aggregation is proposed to address the issues of sparse distribution,similar local features,and large spatial span of associated points in hazardous chemical cargo point clouds.Firstly,a feature aggregation method is used to extract the correlation features between point clouds in space,solving the problem of similar local features.Then,a multi head attention module is used to extract the correlation information between features,The problem of sparse distribution and large span of association points was solved.Finally,attention bias method was used to improve the classification performance of the model.The results showed that the feature extraction and classification algorithm proposed in this paper has the highest classification accuracy compared to existing algorithms.(3)A model transfer method based on contrastive learning is proposed to address the issues of multiple types of hazardous chemical storage,large differences in warehouse size and type,and poor model transfer classification performance.The method uses unsupervised contrastive learning to focus on extracting the features of the data itself,improving the transfer performance of the model when the target domain category label does not exist in the source domain dataset.At the same time,an inter domain adapter is added,By training the network in the target domain with a small amount of data,the classification performance of the migrated model has been significantly improved.The experimental results show that our algorithm has the highest classification accuracy compared to existing algorithms in Zero shot and N-shot migration tasks.