Volume 14, Nº 3 (2023)
Recognition of digital sequences using convolutional neural networks
Resumo
The relevance of identifying tabular information and recognizing its contents for processing scanned documents is shown. The formation of a data set for training, validation and testing of a deep learning neural network (DNN) YOLOv5s for the detection of simple tables is described. The effectiveness of using this DNN when working with scanned documents is shown. Using the Keras Functional API, a convolutional neural network (CNN) was formed to recognize the main elements of tabular information — numbers, basic punctuation marks and Cyrillic letters. The results of a study of the work of this CNN are given. The implementation of the identification and recognition of tabular information on scanned documents in the developed IS updating information in databases for the Unified State Register of Real Estate system is described.
Program Systems: Theory and Applications. 2023;14(3):3-36
3-36
Analysis of images of a plant obtained from a camera of an automated care system to visually assess the change in its state over time
Resumo
A method is proposed for analyzing plant images obtained from a single camera to determine the crown of a plant and detect its individual color shades. The possibility of visual assessment of the state of plants is considered. A description is given of the experimental installation for automated plant care, with the help of which the analyzed data were collected.
Program Systems: Theory and Applications. 2023;14(3):37-58
37-58
Computing of umls concepts etiopathogenetic image using graph metrics
Resumo
At present, the development of clinical decision support (CDS) tools is a crucial task in medical informatics. A lot of different information searching algorithms are used in CDS systems. A fundamental step in the design of these algorithms is the creation of an etiopathogenetic image for the analysis of unstructured medical texts. In this paper, we have conducted the literary review and a comparative evaluation of analytical metrics used to compute the etiopathogenetic image of concepts within the graph model of the Unified Medical Language System (UMLS) metathesaurus. Subsequently, we developed and validated our version of a graph metric suitable for the aforementioned task implementation.
Program Systems: Theory and Applications. 2023;14(3):59-94
59-94
Using of neural networks to search for errors of patient's positioning on chest X-rays
Resumo
The paper presents the results of the application of transfer learning of deep convolutional neural networks for the task of searching for chest X-rays with errors of patient styling and positioning. Evaluated neural network architectures: InceptionV3, Xception, ResNet152V2, InceptionResnetV2, DenseNet201, VGG16, VGG19, MobileNetV2, NASNetLarge. For training and testing we used chest X-rays from open datasets and the unified radiological information service of the city of Moscow. All the models obtained had diagnostic accuracy metrics above 95., while models based on the ResNet152V2, DenseNet201, VGG16, MobileNetV2 architectures had statistically significantly better metrics than other models. The best absolute values of metrics were shown by the ResNet152V2 model (AUC =0.999 , sensitivity=0.987, specificity=0.988, accuracy=0.988, F1 score = 0.988). The MobileNetV2 model showed the best processing speed of one study ($67.8 pm5.0$ ms). The widespread use of the algorithms we have obtained can facilitate the creation of large databases of high-quality medical images, as well as optimize quality control when performing chest X-ray examinations.
Program Systems: Theory and Applications. 2023;14(3):95-113
95-113
Multichannel non-blocking system area network with direct channels
Resumo
Three new versions of a non-blocking self-routed network with direct channels have been developed, in which packet collisions are detected and resolved at the entrance to the network. Collision-blocked packets are retransmitted by sources with minimal delay. End-to-end feedback loops have been added to the network to quickly detect packet collisions. The presence of feedback loops allows for arbitrary traffic without packet loss. Networks are developed in 1-, 2-, 4-, and 8-cascade variants with scaling the number of channels from several hundred to many millions with an assessment of the network speed achieved on permutation traffic. The practical orientation of the proposed networks is system networks with the transmission of routing information in packet headers using control routing information in each cascade. The proposed networks are made in an extended circuit basis, consisting of switches, multiplexers and demultiplexers. The paper presents the characteristics of the constructed networks and gives their comparative assessment. The novelty of the developed networks lies both in the very fact of their construction, and in the structure of the networks used and their characteristics.
Program Systems: Theory and Applications. 2023;14(3):115-138
115-138