ARTIFICIAL INTELLIGENCE-SUPPORTED ANOMALY DETECTION SYSTEM IN UNDERWATER MAPS USING SONAR SYSTEMS

Abstract

The invention relates to a system that works with sonar systems and performs artificial intelligence-based analyses for mapping underwater structures and environmental features. Specifically, the integration of an AI-powered deep learning model that enables the collection and processing of underwater data and the detection of anomalies using sonar data provides a technological framework that makes it possible to detect and map underwater objects, terrains, and biological entities for use in fields such as maritime, environmental monitoring, and defense.

Description

1 TARIFF ARTIFICIAL INTELLIGENCE-ASSISTED UNDERWATER MAPS WITH SONAR SYSTEMS ANOMALY DETECTION SYSTEM TECHNICAL AREA 5 The invention is for mapping underwater structures and environmental features. a system that works with sonar systems and performs artificial intelligence-based analyses It is relevant. In particular, the collection and processing of data relating to the underwater environment and sonar. AI-powered systems that enable anomaly detection by utilizing data. Integration of deep learning models into 10 sectors such as maritime, environmental monitoring and defense. for use in the fields of underwater objects, terrain and biological entities It offers a technological framework that makes its detection and mapping possible. PREVIOUS TECHNIQUE Current systems used in underwater mapping and anomaly detection, 15 mostly multibeam sonar, side-scan sonar, It is based on LIDAR and photogrammetry technologies. These systems measure the seabed. mapping the morphology of underwater shipwrecks in three or two dimensions, to detect anomalies such as structural deterioration or environmental changes It is used. However, there are some technical limitations in the application of these technologies. 20 These include high equipment and operating costs, turbidity in the underwater environment, and light. insufficiency and sensitivity to environmental factors such as thermocline in the data processing process. High processing power requirements and the risk of false positive / negative detection make existing systems difficult. It reduces its efficiency. In recent years, artificial intelligence has been used to increase the effectiveness of these systems. Technologies, particularly deep learning methods, are integrated into underwater data analysis. Deep learning algorithms have begun to utilize data obtained from sonar and sensors. It can automatically and in real-time analyze large volumes of data, thus eliminating the need for manual analysis. This reduces the time loss and risk of errors associated with the evaluation process. Furthermore, this... The systems can withstand challenging environmental conditions such as low resolution, echo, and insufficient light. It possesses sophisticated object recognition and anomaly detection capabilities. Thus, underwater Observation systems are becoming more sensitive and stable. However, underwater also monitor environmental changes that occur over time using maps It is important. Fluctuations in water temperature, sediment transport, and biological factors. 2 activities, industrial pollution, agricultural waste and geological events (e.g. earthquakes or Factors such as landslides cause sudden or gradual changes in underwater morphology. This enables us to open up, which in turn ensures navigation safety, infrastructure stability, and ecosystem health. This poses a risk. Therefore, a regular and detailed examination of the underwater environment is necessary. systems for monitoring in this way, early warning mechanisms and environmental 5 This is of great importance in terms of evaluation. THE PURPOSE OF THE INVENTION The invention improves upon existing systems used in underwater mapping and anomaly detection. artificial intelligence and deep learning developed to overcome its technical limitations 10 It is related to a supported system. This system derives large amounts of data from sonar data. It enables the rapid, accurate, and reliable analysis of large volumes of data; In this way, it increases operational efficiency and prevents potential problems in underwater environments. It enables more effective detection of structural and environmental anomalies. Traditional methods require a large amount of data for the analysis of underwater structures. They are reviewed manually, and this process is both time-consuming and prone to errors. It involves risk. The artificial intelligence-assisted analysis used in the system that is the subject of the invention. The module eliminates this problem by processing sonar data in real time. It eliminates; significantly shortens analysis time and accelerates decision-making processes. It is profitable. 20 In addition, thanks to deep learning algorithms, the system can operate at low resolutions. It can make highly accurate detections even in sonar images that contain uncertainty. and minimizes the rate of human error. This is beneficial both scientifically and otherwise. more reliable in research as well as military or industrial applications. It enables the achievement of results. 25 However, the developed system can be used in different application areas. It has flexibility. Submarine detection, mine detection, protection of underwater infrastructure, many strategic applications such as cable and pipeline deformation monitoring and port security It can be used effectively in the field; thus, in maritime, environmental engineering and It offers an innovative solution in various disciplines, such as the defense industry. 30 In conclusion, the described invention is for underwater mapping and anomaly detection. by solving technical problems encountered in technologies, making them faster, more accurate and comprehensive. It offers a monitoring and analysis capability, and in this respect, it goes beyond existing techniques. It demonstrates performance. 3 LIST OF FIGURES Figure 1. Figure 2. The corresponding numbers in the figures are: 5 1a. Periodic or real-time underwater mapping data from sonar systems. timely collection from the field 1b. Obtained by processing underwater mapping data from different time periods. to be done 1c. Neuron 10 1d. Input vector 1e. Hidden layer 1f. Encoder 1g. Hidden knot 1h. Solvent 15 1i. Reconstructed input vector 1j. Trained model 1k. Defining the error threshold value. 2a. New input data 2b. Reconstructed data 20 2c. Error value determination. 2d. Classification DETAILED DESCRIPTION OF THE INVENTION The invention provides for the periodic or actual detection of data from the underwater environment via sonar systems. mapping data collected over time is processed by an AI-powered autoencoder. It relates to a system that enables anomaly detection by analyzing it using an algorithm. This system is particularly useful when considering seabed topography, biological activity, or potential It is designed for monitoring threats. In the first phase, underwater Raw data for mapping was collected from the field via sonar devices. is collected (1a). These data show the changes in a particular region over time. It is obtained and processed by repeating the process at different times for examination. is converted into usable data (1b). 4 The invention initially involved underwater mapping using a sonar device. raw data is collected from the field periodically or in real time to generate data. data is collected (1a). The sonar device detects underwater structures and the surface a device that emits acoustic waves to determine the topography of the underwater environment, and these waves... a measurement that analyzes the echoes obtained by reflecting off a surface and returning 5 It is a system. This device sends sound waves that bounce back from target surfaces. by calculating their durations and densities, a detailed map of the seabed. These devices are able to generate waves based on the principles of wave propagation and echo reception. Regarding the location, size, and structural characteristics of natural or artificial objects underwater. It enables the precise collection of data. Within the scope of the invention, 10 The sonar device used is also equipped with one of these technologies, and the accuracy of the data... and is of critical importance in terms of resolution. The aforementioned underwater mapping Raw data obtained for this purpose are processed to create a surface map. This data is used for specific purposes. by repeating the process at different time intervals to monitor surface conditions in a region is collected (1b). 15 Autoencoders consist of three parts: encoder (1f), decoder (1h) and hidden node (1g). It consists of a main component. Autoencoders are artificial intelligence that mimics the functioning of the human brain. It works using neural networks. Artificial neural networks are also similar to the human brain. It consists of neurons (1c). The learning process takes place in neurons (1c). This happens. During learning, these networks consist of many user-defined 20 It has a hidden layer (1e). As the number of hidden layers (1e) increases, the learning process It is getting even deeper. The hidden layer receives data from the input layer. By transforming it through weighting and activation functions, the network becomes more It is an intermediary layer that enables the learning of complex and abstract features. These layers, It improves the accuracy of pattern recognition and classification processes in data, 25 It forms one of the fundamental building blocks of deep learning architectures. The part of the autoencoders from the input vector (1d) to the hidden node (1g) region It is expressed as the encoder (1f) section. Here, by taking the input data into consideration The hidden node obtained from the encoder, (1.1) ! !" # $%&! "= It is expressed as follows. Here, z is the hidden node vector, he is the encoder (1f), x is the input It shows the vector (1d) and the encoder (1f) parameters of E. Similarly, The part of the reconstructed input data (1i) up to the hidden node (1g) solver This is section (1h). In this section, the reconstructed input vector is (1i) (1.2) It is expressed as follows. Here, the reconstructed input vector (1i), hD represents the solvent (1h) and D represents the solvent (1h) parameters. When training in autoencoders, the encoder (1f) and decoder (1h) parameters are 10 By minimizing E and D, (1.3) The equation is obtained. Here, the encoder (1f) and solver (1h) vectors are 15 the difference between the optimum encoder (1f) and decoder (1h) respectively It shows the parameters. The optimum encoder and decoder obtained. An autoencoder model is created with its parameters. Different aspects of the training model... Model trained using underwater mapping data at measurement times (1j) is created. Using the trained model (1j) and the error values ​​in the training model, 20 The error threshold value is determined at a 95% confidence interval (1k). This collected data is given as input to the artificial intelligence system, where it is used extensively. A layered artificial neural network architecture is used. In this architecture, each layer performs information processing. Layers consisting of neurons with capacity (1c) from the input vector (1d) Starting with the hidden layers (1e), it processes the data. The encoder part (1f) processes this 25 This process involves transforming the input vector into a smaller, more meaningful representation. As a result, the data is compressed into a structure called a hidden node (1g). This structure stores the basic information in the input data in a summarized form and is transferred. It increases processing speed by reducing the system's data density. This compressed data, held in the hidden node (1g), is then processed by the decoder unit (1h) 30 It is processed to be restored to its original dimensions. This process ! ! !" # $%&! "= ! ! ! ! ! " # # ! " ! "!! "! #$% &'" '(= ! !"# "$! !! " #! "! 6 As a result, the reconstructed input vector (1i) is created and the system data The extent to which it can be restructured correctly is being evaluated. This process As a result, the trained model (1j) created with the training data, the new incoming data It is prepared to be interpreted. The accuracy of the trained model (1j) is obtained It is measured through the error value, which is the difference between the output and the original input; this 5 In this way, the system distinguishes between normal and abnormal data by setting an error threshold value (1k). It creates the groundwork for doing so. After the training process is complete, the system will receive new input that has not been used in training. By feeding the data (2a) into the model, it produces reconstructed data (2b). New input The difference between the data (2a) and the reconstructed data (2b) is 10 as the error value (2c). is calculated. This error value (2c) is calculated with the previously determined threshold value. The data is compared, and if the error is above this threshold, the system labels the dataset as "abnormal". It classifies it as (2d), otherwise it considers it as “normal”. Thus The system uses underwater data that differs from the training data and could potentially be risky. It can detect changes. 15 As a result, this invention enables the continuous and accurate acquisition of underwater mapping data. It is possible to monitor automatically, detecting sudden changes or potential Early detection of threats is ensured. The system is used in environmental monitoring, military defense, and water. as an effective monitoring and decision support tool in many areas, such as infrastructure audits. It has the capacity to be used. In addition, the AI-based classification structure 20 This minimizes the time loss and error rate resulting from manual data analysis. a more reliable and scalable data evaluation infrastructure is being developed. is provided. 30

Claims

7 REQUESTS 1. Periodic or real-time monitoring of the underwater environment via sonar systems. anomaly detection from mapping data collected over time It is a system that provides; 5 features. - Collecting raw data from the field for underwater mapping. sonar devices that make it possible, - A multilayer artificial neural network to which this collected data is given as input. using its architecture; 10 Each of which is a neuron with the capacity to process information (1c) layers, Starting from the input vector (1d), the hidden data is processed through it. layers (1e), It transforms the input vector into a smaller, more meaningful representation, and As a result of this process, the data is sent to a node called the hidden node (1g), entry 15 It stores the basic information in the data in a summarized form and organizes the system's data. an encoder that converts the data into a structure that increases processing speed by reducing its density Part (1f), The aforementioned compressed data is held in the hidden node (1g) original 20 that works to create the input vector (1i) by returning it to its dimensions Solvent (1h) unit, Trained systems, built with training data, interpret incoming data. model (1j), To be used in determining the accuracy of the trained model (1j), Error 25 is used to create a basis for distinguishing between normal and abnormal data. threshold value (1k), By providing new input data (2a) that has not been used in training to the model generated reconstructed data (2b), The difference between new input data (2a) and reconstructed data (2b) containing the error value (2c) and the threshold value (1k) 30 with error value (2c). by comparison, if the error is above this threshold, the system data set is reset. classify as “abnormal” (2d), otherwise considered “normal” by including an artificial intelligence system configured to do so It is characteristic. 8 2. Periodic or real-time monitoring of the underwater environment via sonar systems. anomaly detection from mapping data collected over time It is a method that provides the following feature: - Raw data for underwater mapping is obtained using sonar devices. (1a), 5 - This collected data is processed using a multilayer artificial neural network architecture. given as an introduction to artificial intelligence systems - Each consists of neurons (1c) that have the capacity to process information. layers, starting from the input vector (1d) to hidden layers (1e) processing data through, 10 - The smaller dimension of the encoder (1f) part of the input vector, transforming it into a meaningful representation and, as a result of this process, making the data confidential transfer into a compressed structure called a node (1g), - Compressed data held in the hidden node (1g) is then decoded (1h) to be processed by the unit and returned to its original dimensions and 15 constructing the reconstructed input vector (1i), - Errors to create a basis for distinguishing between normal and abnormal data. Determination of the threshold value (1k), - By giving the model new input data that has not been used in training (2a) again generated constructed data (2b), 20 - Difference between new input data (2a) and reconstructed data (2b) Comparison of the calculated error value (2c) with the threshold value (1k) and If the error is above this threshold, the system marks the dataset as "abnormal". classification (2d), otherwise considered “normal” classification It is characterized by its steps. 25

Images