Page 233 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

224 Chapter 8 A review on plant diseases recognition through deep learning

Table 8.4 State-of-the-art deep learning models [49].

Deep learning
S.no models Parameters Key features and Pros/Cons
1. AlexNet 60 M It was the initial contemporary CNN model, acted as the best image
recognition model. To avoid overfitting, the dropout technique is
employed.
2. ZFNet 42.6 M Improved accuracy with a reduction in weight (7 7 kernels)
3. LeNet 60 k Initially constructed CNN model with limited computational performance.
4. OverFeat 145 M The first single CNN model, classifies, detects and localize objects with
many parameters when compared to AlexNet.
5. GoogLeNet 7 M Improved accuracy at its time with minimum parameters when compared
to AlexNet.
6 VGG 133e144 M It uses a large number of parameters and hence computationally complex
and expensive.
7 DenseNet 7.1 M Better accuracy with minimum parameters and dense connections
8. ResNet 25.5 M Enhanced performance when compared to GoogLeNet and VGG.
Addresses vanishing gradient problem.
9. Xception 22.8 M Depth-wise separable convolution model produce better accuracy than
Inception-v3, ResNet and VGG
10. SqueezeNet 1.25 M Uses 1 1 filter with 50 times fewer parameters than AlexNet with
large activation maps of convolution layers
11. VGG-Inception 132 M Combined version of VCC and inception model with 5 5 convolution
along with two 3 3 layers. Accurate testing model than other most
of the DL models.
12. MobileNet 4.2 M Depth-wise separable convolution concept is used here with nearby
accuracy of GoogLeNet and VGG.
13. Modified/ 0.5/0.54 M Minimum parameters than MobileNet while providing the same accuracy
Reduced
MobileNet

• Obtain a quantitative proportion of the measures of the virus
in the sap of contaminated plants and sanitized suspensions.
• Examine the centralization of the infection in various phases of
advancement of the plants.
• Recognize and examine relationships between infections.
• The antisera can be stored for the testing of suspected plants
later.
• It spares the time lost by brooding.