WiMi Hologram Cloud Inc (WIMI)

[StockpickerCAPC]

WIMI (NASDAQ: WIMI) develops a bionic pattern recognition (BPR)-based convolutional neural network (CNN) image classification technology solution.

Source
https://cj.sina.com.cn/articles/view/7651844612/1c815e20402001gjlu?from=finance

June 5, 2023

In recent years, with the continuous development and application of artificial intelligence technology, image classification technology has been widely used in many fields. And with the rise of deep learning, convolutional neural network (CNN) has become the mainstream model for processing image classification tasks. CNNs recognize images by automatically extracting features from them, and classify them using the softmax function. However, due to the limitations of the softmax function, traditional CNN models have some deficiencies in image classification.

It is reported that in order to solve this problem, WIMI has developed a new image classification method that uses a hierarchical structure inspired by the animal visual system to automatically extract features from images. This method combines bionic pattern recognition (BPR) with CNN, which can make full use of the geometric structure of high-dimensional feature space, so as to achieve better classification performance, so it can overcome some shortcomings of traditional pattern recognition. The method has been validated in multiple experiments and, in most cases, achieves higher classification performance than traditional methods.

A Convolutional Neural Network (CNN) is a deep learning model specifically designed to process images. It can automatically extract features from images through convolution and pooling operations, and use fully connected layers for classification. The convolution operation refers to applying a convolution kernel (also known as a filter) to each location on the image and outputting the result as a feature map. The pooling operation refers to downsampling on the feature map to reduce the amount of computation and the risk of overfitting.

In the traditional CNN image recognition classification model, the softmax function is used for classification. The softmax function can convert a set of scores into a probability distribution, where each score represents a confidence score that the image belongs to a certain class. Traditional pattern recognition methods usually use hyperplanes in feature space to segment categories. However, this method has some disadvantages, such as requiring manual feature selection and difficulty in handling nonlinear data. In contrast, bioinspired pattern recognition (BPR) can overcome these problems by performing class recognition on geometric cover sets that are unioned in a high-dimensional feature space.

BPR is a pattern recognition method based on bionics.
Its basic idea is to use biological systems to simulate the processing of sensory information, and regard the pattern recognition process as being carried out in a high-dimensional feature space. In this high-dimensional space, each sample point is regarded as an object rather than a point. Therefore, samples of different classes are distributed in different regions in the high-dimensional feature space, and these regions are called geometric covering sets. Each geometry cover set consists of a set of geometric objects called geometric primitives, such as spheres, cones, polyhedra, etc. With a proper combination of geometric primitives, a coverage set with high classification performance can be constructed to achieve class recognition.

Research shows that WIMI combines BPR and CNN to achieve better image classification effect. Specifically, based on bionic pattern recognition (BPR) convolutional neural network (CNN) image classification, CNN features can be mapped into a high-dimensional feature space, and a geometric coverage set can be constructed in this space, and then the new sample map Go into that space and decide what category it belongs to.

According to the data, WIMI uses a mapping function to map CNN features into a high-dimensional feature space in BPR-based CNN image classification. This mapping function can be a simple nonlinear transformation such as a polynomial transformation or a radial basis function (RBF) transformation. It is also possible to use some more complex function, such as a neural network or a support vector machine (SVM), to learn this mapping function, converting the CNN features into a form that is easier to classify in a high-dimensional feature space.

WIMI Holographic CNN-BPR image classification technology uses proven geometric primitives with high classification performance in high-dimensional feature spaces, such as spheres, cones or polyhedrons, to construct geometric coverage sets. Then, we can use some optimization algorithm, such as genetic algorithm or particle swarm optimization algorithm, to search for the optimal combination of geometric primitives to construct the optimal geometric cover set. Finally, we can use a classifier, such as the K-Nearest Neighbors algorithm or a Support Vector Machine (SVM), to identify the class to which the new sample belongs.

The specific way to realize the image classification method combining BPR and CNN is as follows:
Prepare training and test datasets:
You need to collect a dataset that contains images of many different categories.
This dataset should consist of two parts:
a training dataset and a testing dataset. The training data set is used to train the CNN model, and the test data set is used to test the performance of the classifier.
Train the CNN model to extract image features:
Use the training dataset to train the CNN model and use the model to extract the features of each image. These features will be used to construct a geometric cover set in a high-dimensional feature space.

Mapping CNN features into a high-dimensional feature space:
A mapping function needs to be used to map CNN features into a high-dimensional feature space. This mapping function can be learned using some nonlinear transformation, such as polynomial transformation or RBF transformation, or using more complex functions, such as neural network or SVM.
Build a geometric cover set:
Use some geometric primitives that have been proven to have high classification performance in high-dimensional feature spaces, such as spheres, cones, or polyhedra, to build geometric cover sets. Then, we can use some optimization algorithm, such as genetic algorithm or particle swarm optimization algorithm, to search for the optimal combination of geometric primitives to construct the optimal geometric cover set.
Classify new samples:
Finally, use a classifier, such as K-Nearest Neighbors or SVM, to identify the class to which the new samples belong. We can map the features of a new sample into a high-dimensional feature space, then find the nearest geometric cover set in this space, and finally classify the new sample into the category represented by the cover set.


In addition, WIMI's CNN-BPR image classification technology features a combination of convolutional neural network and bionic pattern recognition, and performs image classification by constructing a geometric coverage set in a high-dimensional feature space. Compared with the current traditional CNN model that uses the softmax function for classification, the softmax function has limited capacity and cannot handle complex classification problems well, such as image classification. In addition, CNN models cannot fully exploit the geometric structure of high-dimensional feature spaces, thus failing to achieve optimal classification performance. And traditional pattern recognition methods usually need to manually select features and classifiers, requiring a lot of manpower and time costs.

By combining BPR and CNN, this technology can overcome some shortcomings of traditional pattern recognition, improve the performance of image classification, and can handle complex image classification problems. This method can overcome some shortcomings of the current traditional pattern recognition in image classification and in most cases, it has higher classification performance than traditional methods. And it can handle complex image classification problems, such as image recognition, object detection and image segmentation.

At present, image classification technology based on convolutional neural network has been widely used in many fields. The method of WIMI hologram combined with bionic pattern recognition can overcome the limitations of traditional pattern recognition methods and improve the accuracy and reliability of image classification. It is believed that with the continuous development and progress of technology, this technology will have wider application and better performance in the future.