O‘PKA TOMOGRAFIK TASVIRLARIDA SARATON KASALLIGINI TURLARI BO‘YICHA TASNIFLASHNING VEYVLETLARGA ASOSLANGAN KONVOLYUTSION NEYRON TARMOQ MODELI Zaynidinov Hakimjon Nasriddinovich, Jurayev Umidjon Sayfullayevich Texnika fanlari doktori, professor Guliston davlat universiteti Qisqacha mazmuni: Ushbu maqolada biz o‘pka tasvirlarini ajratib
olish va tasniflash uchun veyvlet konvolyutsion neyron tarmog‘i (WCNN)
arxitekturasini taklif qilamiz. WCNN modeli o‘pka tasvirini tasniflashning aniqligi
va samaradorligini oshirish uchun konvolyutsion neyron tarmoqlari va veyvlet
o‘zgartirishlarning kuchli tomonlarini birlashtiradi. Natijalar WCNN modelining
o‘pka tasvirlarini tasniflash samaradorligini ko‘rsatadi, bu o‘pka saratoni tashxisi
va davolashning aniqligini oshirishi mumkin.
Taklif etilayotgan model to‘rtta asosiy
bosqichdan, ya’ni dastlabki ishlov berish, xususiyatlarni ajratib olish, tasniflash va
vizuallashtirishdan iborat. Oldindan ishlov berishda o‘pka tomografik tasvirlaridan
tuz-qalampir shovqinini olib tashlash uchun median filtr qo‘llanildi. Diskret veyvlet
o‘zgartirishni amalga oshirish uchun Dobeshi veyvlet o‘zgartirishi qo‘llanilgan.
Taklif etilayotgan modelda past darajadagi tafsilotlarni olib tashlash va tasvirlar
hajmini kamaytirish uchun tasvirlarga 3-darajali Dobeshi veyvlet parchalanishi
qo‘llanilgan. Natijalar shuni ko‘rsatadiki, tavsiya etilgan usul 97% aniqlik bilan
yaxshi natijalar beradi.
Kalit so‘zlar: veyvlet,
raqamli ishlov berish, neyron tarmoq, chuqur o‘qitish,
Dobeshi, konvolyutsion neyron tarmoq, filtrlash, median, tomografik tasvir.
Abstract: In this paper, we propose a wavelet convolutional neural network
(WCNN) architecture for lung image extraction and classification. The WCNN model
combines the strengths of convolutional neural networks and wavelet transforms
to improve the accuracy and efficiency of lung image classification. The results
show the effectiveness of the WCNN model in classifying lung images, which can
improve the accuracy of lung cancer diagnosis and treatment. The proposed model
consists of four main steps, namely pre-processing, feature extraction, classification
and visualization. A median filter was used in preprocessing to remove salt-and-
pepper noise from lung tomographic images. Dobeshi wavelet transform was used
to perform discrete wavelet transform. In the proposed model, 3rd-order Dobeshi
wavelet decomposition is applied to images to remove low-level details and reduce
image size. The results show that the proposed method gives good results with 97%
accuracy.
Keywords: wavelet, digital processing, neural network, deep learning,
Dobeshi, convolutional neural network, filtering, median, tomographic image.