Acceso abierto

Research on the optimization method of image classification model based on deep learning technology and its improvement of data processing efficiency

Yi Zhang

Yi Zhang

Jiangsu Police InstituteNanjing, China
Buscar este autor en
Sciendo|Google Scholar
Zhang, Yi
  
19 mar 2025
Research on the optimization method of image classification model based on deep learning technology and its improvement of data processing efficiency's Cover Image
Acerca de este artículo
Artículo anterior
Artículo siguiente
Cite
Descargar portada
Publicado en línea: 19 mar 2025
Recibido: 11 nov 2024
Aceptado: 20 feb 2025
DOI: https://doi.org/10.2478/amns-2025-0395
Palabras clave
© 2025 Yi Zhang, published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1.

The basic structure of the convolution neural network
The basic structure of the convolution neural network

Figure 2.

Performance of different network models
Performance of different network models

Figure 3.

Error comparison of different network models
Error comparison of different network models

Figure 4.

Classification of models
Classification of models

The classification effect of different models is compared

Model Weighted accuracy Loss Mean standard deviation Training time(Mine)
VGG16 0.9683 0.7596 0.0254 223
ResNet18 0.9714 0.7651 0.0198 215
MobileNetV2 0.9723 0.8123 0.188 322
DenseNet 0.9713 0.7652 0.175 215
EfficientNetB 0.9821 0.7625 0.115 315
Ours 0.9823 0.7023 0.008 135

Different network model parameters

Network model Network number Input size Parameter number FlOPS(G)
VGG16 16 224×224×4 135.5 16.15
ResNet18 18 224×224×4 11.5 2.05
MobileNetV2 54 224×224×4 3.2 0.54
InceptionV3 48 299×299×4 22.1 6.84

The training situation of different models is compared

Network model Network number FlOPS(G) Accuracy rate Training time Test time
VGG16 16 16.15 0.945 43min37s 0.0295s
ResNet18 18 2.05 0.932 29min45s 0.0262s
MobileNetV2 54 0.54 0.948 27min53s 0.0299s
InceptionV3 48 6.84 0.966 30min43s 0.0301s

The impact of different learning rates on model performance

Optimizer learning_rate batch_size smooth_val epoch Weighted accuracy Training time
RMSprop 0.002 16 0.1 100 0.9547 1h55m
0.008 16 0.1 100 0.9658 1h56m
SGD 0.002 16 0.1 100 0.9352 1h48m
0.008 16 0.1 100 0.9763 1h47m
Ours 0.002 16 0.1 100 0.9985 1h25m
0.008 16 0.1 100 0.9992 1h31m
Idioma:
Inglés
Calendario de la edición:
1 veces al año
Temas de la revista:
Ciencias de la vida, Ciencias de la vida, otros, Matemáticas, Matemáticas aplicadas, Matemáticas generales, Física, Física, otros
RSS Feed de revista