В настоящее время остаётся актуальным интеллектуальный анализ электрокардиограммы (ЭКГ) в нейронных сетях, классификация болезней по результатам анализа, определение вида болезни на основе признаков, изучение и установление диагноза причин возникновения болезни. В данной статье изучено влияние форматов фото при классификации ЭКГ сигналов в сетях CNN, сделан анализ форматов фото и изменений, исследованы процессы разработки фото (изменение размера, сжатие), их влияние на точность и качество работы сети.

Электркардиограммани (ЭКГ) нейрон тармоқларида интеллектуал таҳлил қилиш, таҳлил натижаларига қараб касалликларни таснифлаш, белгиларига асосан касалликни турини аниқлаш, келиб чиқиш сабабларини ўрганиш ва ташҳис қўйиш бугунги кунда долзарб бўлиб қолмоқда. Ушбу мақолада CNN (convolutional neural network) тармоғида ЭКГ сигналларини таснифлашда тасвир форматларининг таъсири ўрганилди. Шунингдек, тасвирларнинг формати ва формат ўзгаришлари, тасвирларга ишлов бериш жараёнларлари (ўлчамини ўзгартириш, сиқиш) тармоқ аниқлигига ва тармоқнинг иш сифатига таъсир ўтказиши таҳлил қилинди.

Data mining of electrocardiography in neuronal networks, classification of diseases according to the results of analysis, definition of the type of disease, the causes of its origin and diagnosis are still topicality today. In this article has been examined the effects of image formats for classifying ECG signals in convolutional neural network, CNN. It also analyzed the formatting of images and format changes, image processing (resizing, compression) on the network accuracy and the network performance. Classifying of ECG images in different formats is performed on AlexNet and GoogleNet networks. Both networks’ opportunities are defined and compared. In this experiment is also used ECG data of three different groups of people: those who have heart arrhythmia (ARR), those with stable heart failure (CHF) and those who have healthy heart function (NSR). The total number of data is 162, which is derived from the three sections of the PhysioNet database. Additionally, there were 96 cardiac arrhythmias, 30 stable heart failure and 36 healthy heart function.Artificial neural networks are widely used to identify the cause of the disease in the medical field and improve the diagnostic process. In particular, world wide research is being carried out on diagnostics and diagnosis depending on functional changes of cardiovascular system. CNN networks show 99.9% of the automatic diagnosis. In contrast to the normal picture classification of electrocardiography classification, the usual CNN networks are mainly selected for high quality images. As a result, the new image rendering tests will affect the network’s classification capability as a result of re-processing, such as compression and resizing the image quality.In CNN networks, it is often compressed into a very small number of images in any format, in other words, usually any image is compressed into the network.Therefore, the format of the preprocessing images is important, which means that the image is durable or not durable for the in compression process. The CNN requires all the images to be of the same size. Therefore, the result of the network depends on the size of the picture and how it should be changed.Experience has revealed that the formatting of AlexNet network and the accuracy of electrocardiography classification are more stable than the GoogleNet network. In the classification of ECG signals, the network has been more successful than other formats when working with JPG formats. This difference was especially noticeable in GoogleNet network.