Page 196 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 6 Plant leaf disease classification based on feature selection 185

Table 6.9 Details of selected features.

Before FS After FS
Statistical Mean R, mean G, mean B, mean H, mean S, Mean R, mean G, mean H, mean S, mean V,
features mean V, R standard deviation, G R standard deviation, G standard deviation,
standard deviation, B standard deviation, B standard deviation, H standard deviation, S
H standard deviation, S standard deviation, standard deviation, V standard deviation, R
V standard deviation, R skewness, G skewness, B skewness, S skewness,
skewness, B skewness, H skewness, S R kurtosis
skewness, V skewness, R kurtosis, G
kurtosis, B
kurtosis, H kurtosis, S kurtosis, V kurtosis
Geometric Defect area Defect area
features Defect perimeter Major axis length
Major axis length Eccentricity
Minor axis length Orientation
Eccentricity Convex area
Orientation Equivalent diameter
Convex area Extent
Equivalent diameter Perimeter
Solidity
Ratio
Textural GLCM GLCM
features
After that, this subset of features is used to train the MLP with 50 epochs, which yields the following result. GLCM, gray-level
cooccurrence matrix.

In this approach, MLP achieved 85.2% of training accuracy
and 81.2% of testing accuracy. And with this approach, the
training time is significantly lower than the CNN models
(Tables 6.10e6.12).

5. Conclusion
We have proposed an image-based method with preprocess-
ing steps to identify diseases in mango leaf by using DL. Rescal-
ing, center alignment, and contrast enhancement steps are
used as preprocessing stages, which provide suitable adjustments