Field robot ground classification method and system based on small number of labels

Abstract

The invention discloses a field robot ground classification method and system based on a small number of labels. The invention belongs to the technical field of computers, a vibration signal acquiredby a vibration sensor corresponds to ground image signals acquired by image acquisition equipment. The ground image signals are acquired. Real ground types corresponding to the ground image signals are manually recognized, the vibration signals are labeled, labeled samples and unlabeled samples are obtained to train a support vector machine, the situation that a certain type of samples lack labeling is prevented. Meanwhile, the spatial smoothness hypothesis is better utilized, and the classification accuracy is improved. The sample classification well reflects the actual situation of data classification, a model is trained by using the classified data, and the accuracy of semi-supervised classification is improved.

Description

Technical field [0001] The present invention relates to the field of computer technology, in particular to a ground classification method and system for field robots based on a small number of annotations. Background technique [0002] How to use massive amounts of data is an important task facing machine learning. The traditional support vector machine is a supervised learning method that requires a large number of labeled samples for training. However, in practical applications, since most of the sample data that can be used are unlabeled, there are fewer labeled sample points. If only these few labeled samples are used, it will result in a large number of location-labeled samples. Was lost. Therefore, some scholars have proposed a semi-supervised learning method that uses both unlabeled sample data knowledge and a small amount of labeled sample data knowledge in the process of semi-supervised learning. However, the existing semi-supervised learning methods do not fully utili...

AI Technical Summary

Problems solved by technology

However, existing semi-supervised learning methods do not fully exploit spatial smoothness, especially in robotic ground classification, resulting in a large number of misclassifications

Images