AUTOMATED IMAGE RECOGNITION SYSTEMS USING INSPECTION AND SCREENING COMPLEXES IN CUSTOMS CONTROL PROCESSES

In modern customs operations, the rapid and accurate processing of large volumes of data is crucial, particularly when analyzing images obtained from inspection and scanning complexes (ISCs). Convolutional Neural Networks (CNNs) offer a promising solution, providing enhanced image analysis and classification capabilities. This study focuses on the implementation of a CNN-based algorithm for detecting and marking the contours of firearms in X-ray images from ISCs. The CNN model, developed using the TensorFlow/Keras library, consists of 14 layers, including convolutional, pooling, and fully connected layers. The model was trained on a custom dataset of 150 annotated X-ray images, where data augmentation techniques were employed to improve robustness against geometric distortions and low image quality. The training process involved 300 epochs, and the model's accuracy was evaluated using metrics such as mAP and confusion matrices. The results indicate an 80% accuracy on validation data and an 84% accuracy on training data. The model effectively identifies firearms in diverse images but shows limitations when detecting other firearm types due to the specificity of the training dataset. This research highlights the potential of CNNs in enhancing customs control through automated image recognition, while also emphasizing the importance of diverse training data for improving generalization across different object types.