Training of a Convolutional Neural Network and Its Object Recognition Ability Depending on Illumination and Contrast of Images
| Autor: | Andriy Segin, Dmytro Kyrychuk |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Artificial neural network
Computer science business.industry media_common.quotation_subject Feature extraction ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Cognitive neuroscience of visual object recognition Pattern recognition Feature selection Object (computer science) Convolutional neural network ComputingMethodologies_PATTERNRECOGNITION Contrast (vision) Artificial intelligence business Transfer of learning media_common |
| Zdroj: | ACIT |
| DOI: | 10.1109/acit52158.2021.9548493 |
| Popis: | This paper presents peculiarities of dataset preparation for convolutional neural network training and specificities, which influence the further object recognition by trained network. The influence of illumination and contrast of images from dataset on quality of network training was researched. Rationality of using neural network transfer learning for increasing accuracy of recognition was demonstrated. The object recognition ability of a convolutional neural network while scaling the object or changing objects’ appearance relatively to initial images from training dataset was researched. The results of influence of illumination on quality of object recognition by trained network and influence of background on the quality of feature selection were presented. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|