![Scientific and technical journal «Priborostroenie»](/images/mag-pr.png)
DEVELOPMENT OF A CONVOLUTIONAL LAYER OF A DEEP NEURAL NETWORK FOR DETECTING DEFECTS IN ROLLED METAL
![Scientific and technical journal «Priborostroenie»](/images/mag-pr.png)
Annotation
A convolutional layer of a deep neural network, designed to determine defects in rolled metal products, is considered. To determine the defect, it is proposed to use algorithms for the segmentation of flaw detection images and several types of filtering within the convolutional layer. Filtering is based on the use of combined convolution algorithms with different initial masks. To minimize the error at the output of the convolutional layer, the GELU activation function is used. Results of the experiments are presented.
Keywords
Постоянный URL
Articles in current issue
- AUTOMATIC CONTROL SYSTEM FOR THE MOVEMENT OF INDUSTRIAL PRODUCTS USING RADIO FREQUENCY IDENTIFICATION TAGS
- SYNTHESIS OF SELF-CHECKING BUILT-IN CONTROL CIRCUITS BASED ON LOGICAL COMPLEMENT TO THE EQUILIBRIUM TWO-OUT-OF-FIVE CODE
- OPTIMAL CONTROL OF THE PIEZO ACTUATOR FOR TUNING THE RADIO-REFLECTING MESH OF SPACE REFLECTOR
- ALGORITHM FOR SEQUENTIAL OPTIMIZATION OF THE STAGE OF EXTENDING THE SPOKES OF A LARGE-SIZED TRANSFORMABLE REFLECTOR
- DEVELOPMENT OF A CONVOLUTIONAL LAYER OF A DEEP NEURAL NETWORK FOR DETECTING DEFECTS IN ROLLED METAL
- MULTIPLICATIVE PROCESSING OF THE FREQUENCY SPECTRUM OF THE MOBILE OBJECT RADIATION
- METHOD FOR EVALUATING THE DYNAMIC RESISTANCE OF PIEZOELECTRIC ELEMENTS
- ANALYSIS OF THE PROCESS OF THIN FILM DEPOSITION IN VACUUM CHAMBERS BY HiPIMS METHOD
- RADIO-OPTICAL LENS REFLECTOR-EMITTER
- METHOD FOR DETERMINING THE RELIABILITY OF THE ION CURRENT SENSOR WITH EIGHT HIGH-VOLTAGE PROBES