LIGHTWEIGHT MULTILAYER RANDOM FORESTS CLASSIFIER FOR REAL-TIME OPERATION UNDER LOW-SPECIFICATION AND CLASSIFICATION METHOD USING THEREOF

摘要

The present invention relates to a lightweight multilayer random forest classifier for low-spec real-time operation, and more specifically, as a lightweight multilayer random forest (LMRF) classifier, wherein each layer is a random forest (RF). It is a deep model of a non-neural network type of a layer-by-layer structure composed of, and is characterized in that each layer is composed of a tree of not more than a predetermined number. In addition, the present invention relates to a classification method using a lightweight multilayer random forest classifier for low-spec real-time operation, and more specifically, as a classification method using a lightweight multilayer random forest (LMRF) classifier, (A) each A deep model of a non-neural network type with a layer-by-layer structure in which the layer is a random forest (RF), and each layer creates an LMRF classifier consisting of a predetermined number of trees or less. The step of doing; And (B) performing classification using the generated LMRF classifier. According to the lightweight multilayer random forest classifier for low-spec real-time operation proposed in the present invention and a classification method using the same, each layer is a layer-by-layer structure consisting of a random forest (RF). By constructing a non-neural network type of deep model and configuring each layer into a tree of less than a predetermined number, it provides similar performance to DNN with fewer hyper parameters compared to the existing DNN model, and is used under the same conditions. Since the time processing time is faster than the DNN, it can be applied to the field for real-time processing.