Oleksandr Kravchenko

An automated method for detecting and classifying submicron surface defects on mirrors used in high-precision optical systems has been developed, utilizing interferometric image analysis and deep learning. This approach replaces manual inspection with a neural network, delivering faster and more objective defect diagnostics, eliminating human bias, and accelerating quality control in serial production of optical components. A synthetic dataset of interferometric images was generated to train the model, simulating scratch-type defects on mirror surfaces through specialized software based on the Linnik interferometer model. The dataset encompasses three surface classes: flat surfaces, single scratches, and multiple scratches. The neural network is built upon MobileNetV2, pre-trained on ImageNet, with fine-tuning of its final blocks to adapt to the task’s specifics. The architecture incorporates GlobalAveragePooling2D for feature compression, Dense layers with ReLU activation and BatchNormalization, Dropout to mitigate overfitting, and a Softmax output layer for classifying the three categories. Data augmentation and soft voting techniques were employed to enhance the model’s generalization ability. Classification accuracy, assessed using the accuracy metric, achieves 96% on the synthetic validation set and 82.7% on real images acquired from a Linnik interferometer. The highest accuracy is observed for flat surfaces, while the lowest occurs for multiple scratches, highlighting challenges posed by real-world conditions such as noise and artifacts. The method proves its practical value for automated diagnostics, with future enhancements tied to improving synthetic data realism-potentially by incorporating modeled noise-and extending the model’s adaptability to additional defect types like indentations or protrusions, thereby broadening its applicability in optical system manufacturing

Due to the full-scale invasion of Ukraine, the demining and methods of automating this process are much more relevant nowadays. Robots and algorithms can help solve this problem. In this study, the methods of automated detection of explosive devices were investigated. We reviewed various modern methods of demining the territory. We analyzed the principle of operation of metal detectors and GPR. After that, we found out the advantages and disadvantages of each of these devices. We investigated which of these devices gives more accurate data about the object detected on the ground and which data is better for training the model. As a result, an information system was developed based on a convolutional neural network and an autoencoder for the automated classification of explosive devices. We used a set of scanned ground images obtained from ground- penetrating radar (GPR) for the autoencoder input data. We received an accuracy of 97.83%. The algorithm described in this study can help in demining the territories of Ukraine. Keywords 1 Metal detector, landmines, GPR, machine learning, classification.

Modern information systems are increasingly used in various areas of our life. One of these is the quality control of the condition of the road surface in order to carry out repair work on time if necessary. The machine learning method can facilitate the control process, which was demonstrated in this work. Analyzing the road surface condition using image classification requires much pre-classified data and decent computing power. As the modern need for proper quality control of the road surface is high, it is possible to analyze using sensor-recorded data in tabular form and machine learning methods, which should show high accuracy of the classification results. Development and research of an information system for classifying the condition of the road surface were described in this paper, including ways for optimizing similar approaches and improving the results obtained through the use of a greater number of features, in particular, taking into account not only the speed indicators at the given time of the car's movement but also the performance indicators of internal combustion engine. As a result, an information system was developed that classifies the road surface condition using features obtained from various types of sensors and recorded in tabular form. Machine learning methods such as Random Forest, Decision Tree, Support Vector Method, and AutoML library were used to compare accuracy results using a large set of artificial intelligence methods. The best results were obtained using the Random Forest ensemble machine learning method. The analysis of the classifier according to various parameters was carried out, and a search for the best hyperparameters was performed. At the same time, achieving a 91.9% accuracy of road surface condition classification was possible. Parallel calculations were used during model training. As a result, training time was decreased by 5 times with the use of the CPU and by 51 times with the help of the GPU.