Research on image processing of intelligent building environment based on pattern recognition technology

Abstract

With the continuous development of urbanization, urban population, economy and other factors have a close impact on the geometry and distribution of urban buildings. Obtaining information of urban buildings from aerial images or satellite images quickly and accurately is not only conducive to updating geospatial data, but also of great significance for effective monitoring of new thematic information such as new buildings. Moreover, in recent years, the research and improvement of building recognition and contour extraction algorithms based on satellite images or aerial images are helpful to the recognition and classification of urban buildings. It is of great significance to the acquisition of GIS data, the understanding of images, large-scale mapping and many other applications of remote sensing data. With the development of artificial intelligence and computer technology, the image processing of intelligent building environment based on pattern recognition technology has become an important research direction in the field of intelligent building image recognition. Based on the concept, principle and technology analysis of pattern recognition technology, this paper studies the application of pattern recognition technology in the image processing of intelligent building environment. In this paper, based on image processing of intelligent building as the basic theoretical platform, with the pattern recognition technology as the basic research means, three problems of image processing, image extraction and image recognition in image processing of building intelligent environment are studied respectively, and corresponding reasonable solutions are put forward.

Introduction

Nowadays, with the rapid development of computer hardware and the continuous development of computer applications, people are beginning to require computers to perceive more effectively information such as voice, text, image, temperature, vibration and so on. But in general sense, at present, the general computer can not directly perceive them. Our commonly used keyboard, mouse and other external devices are powerless for this external world. Although cameras, graphics, scanners, microphones and other devices have already solved the conversion of these non electrical signals and connected with computers, but because of the low recognition technology, they failed to make the computer really know what the information was after recording. The low perception ability of computers to the outside world has become the bottleneck of exploiting computer applications, which is also in sharp contrast with their superb computing ability [1]. As a result, pattern recognition, which aims at broadening the application fields of computers and improving their ability to perceive external information, has developed rapidly.

Pattern recognition in artificial intelligence refers to the use of computers to replace human beings or to help people perceive patterns. It is a simulation of human perception of external functions. It is a computer pattern recognition system, that is, to enable a computer system to simulate human perception of external information through the senses, and to recognize and understand the surrounding environment. The applications of artificial intelligence in pattern recognition include remote sensing, biomedical image and signal analysis, automatic nondestructive testing of industrial products, fingerprint identification, text and speech recognition, machine vision pattern recognition and so on.

Based on the analysis of the application status of building energy management system, a design scheme of intelligent energy management system based on multi-system linkage is proposed [2]. Considering the environmental temperature and passenger flow as the influencing factors of building energy consumption, the optimal control strategy of air conditioning system energy consumption in public buildings is studied based on fuzzy control method. Taking a real commercial building as an application object, the energy-saving effect of intelligent control strategy is simulated, and the effectiveness of energy-saving control strategy is verified.

With the increasing demand for building energy consumption and indoor comfort, higher requirements are put forward for more effective facade structure [38], [39], [40], [41], [42], [43]. The facade plays an intermediary role between the exterior and interior of the building. It bears many functions that affect the building performance. Compared with the static structure, the intelligent facade provides better performance and adapts to the changing environmental impact and internal requirements through dynamic adjustment. Such a system is being explored and has been applied. The concept of intelligent facade has existed since the early 1980s, and since then, the technological possibilities of implementing intelligent systems have multiplied. Today, the fourth industrial revolution is based on the implementation of intelligent and networked production facilities. Tahmid et al. [3] have changed their understanding and requirements of system intelligence in view of the current industrial exploration of intelligent technology systems. The purpose of this study is to examine the understanding of intelligent systems in the facade and industry context. This provides a basis for future research on the transferability of strategies. This study provides the terminology, related aspects, current definitions and characteristics of each intelligent building system.

As a traditional edge detection algorithm, Canny operator has been widely used and improved. Canny uses lag thresholds. Different thresholds have a great impact on the detection results, but the number cannot directly reflect the detection results. Wang et al. [4] realized four algorithms by Open CV programming [44], [45], [46], [47], [48], [49]. After programming, the results of detection under different thresholds can be visually displayed, which is convenient for image edge detection.

Aiming at the shortcomings of traditional edge detection algorithm, such as low detection accuracy, poor robustness and poor noise immunity, Liu et al. [5] proposed a multi-scale edge detection algorithm based on Canny. The algorithm not only detects the overall contour of the image edge, but also detects the details of the image, and makes a detailed comparison between MATLAB and several classical edge detection algorithms. Compared with the classical edge detection algorithm, the multi-scale Canny edge detection algorithm has higher accuracy and robustness, and is more robust against Gauss noise. It can effectively extract the edge information of the image.

In computer vision, artificial target recognition is usually the central task, and edge detection is the most commonly used method. In order to suppress the influence of noise on image edge detection and foreign object edge detection, the improved Canny algorithm is applied to artificial object edge detection. Firstly, the algorithm uses adaptive smoothing filter to filter salt and pepper noise, which can better protect the details of the image. Then, the edges of the artificial instrument and the external objects are distinguished by the geometric features of the edges. Experiments on high-speed railway field measurement show that compared with the traditional Canny edge detection algorithm, the algorithm can accurately locate the edge of the device and effectively suppress the edge algorithm caused by noise, ballast and object surface texture [6]. Edge detection is the process of detecting and representing the existence and location of image signal discontinuity. It is the basic transformation of signal to symbol, which affects the performance of subsequent higher level downstream mode analysis. Generally speaking, edge detection has two main steps: filtering, detection and location. In the first step, finding the optimal scale of the filter is an ill-posed problem, especially when single scale is used throughout the image [7]. A multi-resolution description of images that can represent the image features appearing within a certain scale is used. The combination of Gauss filters with different scales can improve the single scale problem. In the second step, the edge detector is usually designed to capture the simple ideal step function in the image data, but the actual image signal discontinuity deviates from this ideal form. The other three deviations from step functions are related to the real distortion in natural images. These types are impulse function, slope function and sigmoid function, which represent narrow-line signal, simplified fuzzy effect and more accurate fuzzy modeling respectively. According to this analysis, four general rules of edge representation based on edge type classification are proposed, namely impulse, step and sigmoid (RISS). In addition, the proposed algorithm performs connectivity analysis (CA) on edge mapping to ensure the removal of small, disconnected edges. The experimental results show that the multi-resolution edge detection algorithm based on edge pattern analysis can effectively improve the accuracy of edge detection and location. In order to extend the proposed algorithm to real-time applications, B Jiang proposed a parallel implementation on graphics processing unit (GPU) and achieved good results [8], [9], [10], [11].

With the continuous improvement of living standards, the number of pets is increasing and pet diseases are constantly occurring. Therefore, the prevention of pet diseases is getting more and more attention. Yang et al. [12] put forward an intelligent pet disease prediction system based on ARM and image recognition. The embedded monitoring device is used to monitor pets, and the improved moving object detection algorithm and Hough transformation based linear detection algorithm are used to record abnormal behaviors. With S3 C2440 as the core chip, the embedded operating system is built to the embedded terminal, and the corresponding programs are written to the remote Web users through wireless transmission technology. PEK reached 81.82%, basically reaching the expected goal.

In order to prevent and control the harm of stored grain pests, it is important to find an effective method for identifying stored grain pests by computer. Aiming at the multi class recognition problem of stored grain pests based on image, Cheng et al. [13]. Proposed a new method for image recognition of stored grain insects based on deep convolution neural network. Compared with the traditional warehouse pest identification method, this method greatly simplifies the data preprocessing process, and the accuracy rate reaches 97.61%, which is obviously superior to the traditional method. Therefore, the identification method of stored grain pests based on deep convolution neural network has high practicability, and has the significance of further research and promotion.

The image recognition technology of transformer in augmented reality environment is studied. In order to solve the problem of transformer image recognition in augmented reality environment, based on the introduction of CNN as a typical deep learning model, Chikmurge et al. [14] proposed an improved convolution neural network (CNN) model based on two parallel structures. The improved CNN is used to classify the images scanned by augmented reality camera and realize transformer pattern recognition. Compared with conventional CNN and SIFT image recognition algorithms, the improved CNN algorithm has lower bit error rate and higher accuracy of transformer image recognition. The simulation experiment verifies the correctness of the method.

Artificial intelligence image recognition has brought revolutionary development for medical image recognition. Based on more than 180,000 color photographs of fundus PACS, a deep learning algorithm was developed for automatic detection of diabetic retinopathy (DR). Under laboratory conditions, the sensitivity and specificity test results were 95.3% and 79.5% respectively. The study used 34,100 color fundus photographs uploaded from the Internet by the Airdoc retinal disease intelligent identification center to verify the accuracy of the algorithm in the real world [15]. The sensitivity and specificity of DR detection were 94.6% and 78.4% respectively. The results show that the deep learning algorithm model copies the experience of ophthalmologists; diagnosis of diabetic retinopathy to the auxiliary diagnostic software, and has a high diagnostic efficiency, which is equivalent to the accuracy of human doctors.

Aiming at the over segmentation problem of traditional watershed algorithm, Cao et al. [16] proposed an improved watershed algorithm for image segmentation, which provided technical support for intelligent recognition of tea. The differential equation model is applied to denoise the image. Then, the Otsu algorithm and watershed algorithm are used to do the two segmentation of the de noised image, and a better denoising effect is obtained. Denoising, clear boundaries of tea and good image quality are beneficial to subsequent segmentation of images. For the first time, Otsu algorithm is used to segment the young leaf images, which improves the integrity of leaf images, and uses watershed algorithm to segment young leaf images. The improved watershed algorithm has a good segmentation effect on the tender leaves in tea image.

Aiming at the limitation of convolutional neural network in natural image recognition, a convolutional neural network algorithm for natural image recognition is proposed, which achieves better recognition accuracy and speed. GPU algorithm is used to accelerate the algorithm, and multi region logistic regression is used to improve the accuracy of image recognition. Finally, the correctness and effectiveness of the improved algorithm is verified in the experimental environment.

Target feature extraction plays an important role in image recognition, including face recognition and character recognition. However, feature extraction usually relies on domain knowledge or prior experience. Convolution neural network (CNN) has attracted the attention of researchers because of its self learning ability, which can automatically extract effective features and realize image recognition. However, the features learned by classical CNNs may often have poor discriminant ability for image recognition. The linear discriminant analysis loss (LD.) is introduced into CNN, and a discriminative deep level feature learning method based on linear discriminant analysis (LDA) for image recognition is proposed by Zhao et al. [17]. Therefore, CNNs can provide discriminant features to improve image recognition performance. Methods A deep CNN for feature extraction was constructed, and the effective features for image classification task were automatically extracted by multi-layer perceptron. LDA is introduced and a new linear discriminant loss function (LDLIFE) is developed through a new form of Fisher criterion. The new LDWEST and SOFTMAX losses are integrated into the joint loss function of deep feature learning. The traits learned can minimize classification errors and achieve inter-class allocation and intra-class compactness [18]. In the learning process, the MINI batch-based average strategy is used to update the class center. Results The experimental results on MINIST and CK+ databases show that the average recognition rate of the algorithm is 99.53% and 94% respectively. This method can achieve 100% recognition rate for some classifications in MNIST and CK+ databases. In this paper, a new discriminant depth feature learning algorithm is proposed for image recognition. Experimental results on different databases show that the method can explicitly realize the learning function between class compactness and inter class separability, and further improve the classification ability of features effectively. Therefore, this method can achieve higher recognition rate than some existing methods. This method does not require additional computational load in the Softmax loss comparison during the test phase.