An Intelligent System for Cucumber Leaf Disease Diagnosis Based on the Tuned Convolutional Neural Network Algorithm

Table 1

Previous studies for plant leaf disease classification.

Author	Preprocessing methods	Segmentation methods	Feature extraction methods	Classification methods	Diseases types	Accuracy result (%)
[2]	—	K-means clustering	Color and shape	SR	Downy mildew, bacterial angular, Corynespora cassiicola, scab, gray mould, anthracnose, and powdery mildew. Total images: 420	85.7
[9]	AR-GAN	Combined GrabCut with SVM	Color, texture, and border features	DICNN	Anthracnose, downy mildew, and powdery mildew	90.7 on raw leaf diseased, 96.1 on lesion images
[10]	—	—	CNN	CNN and VGG-net	MYSV, ZYMV, CCYV, CMV, PRSV, WMV, KGMMV, downy mildew, and healthy. Total images: 9000	93.6 for VGG-net; 86.6 for CNN
[11]	Square crop and square deformation	—	CNN	CNN	MYSV and ZYMV. Total images: 800	94.9
[16]	Gaussian filtering used to blur the image to reduce the noise	K-means clustering	HOG	SVM	Alternaria leaf blight, angular leaf spot, bacterial leaf spot, bacterial wilt, cercospora leaf spot, cucumber mosaic, target leaf spot, powdery mildew, downy mildew, and Phytophthora blight	86
[19]	—	Watershed algorithm	Global-localSVD	SVM	Anthracnose, blight, and downy mildew. Each class includes 100 images	—
[20]	Improves the local contrast and makes infected regions more visible	SHSB	VGG-19 and VGG-M	DT, LR, M-SVM, C-SVM, fine KNN, ESDA, and NN	Angular leaf spot, corynespora, anthracnose, downy mildew, powdery mildew, and healthy	98.08 using M-SVM
[22]	Smoothing filtering used to eliminate noise	—	GLCM	ANN	Healthy, downy mildew, and powdery mildew	80.45
[23]	Data augmentation and contrast enhancement performed	—	HOG, BRISK, and FAST	QSVM	Blight, powdery mildew, conrnyspora, angular leaf spot, anthracnose, and downy mildew. Total images: 1010	93.5
[26]	—	—	MIFS, MRMR, and SFS	SVM, NB, KNN	Cucumber chilling injury classes (normal, lightly chilling, and severely chilling). Also, two classes (normal and chilling).	SFS 90.5, SVM 100
[27]	Data augmentation	—	CNN	CNN	Angular spot, anthracnose, black spot, brown spot, downy mildew, gray mold, powdery mildew, and target spot	93.75
[28]	Flip horizontally, vertically, and rotate images	Combined color feature with region growing	—	DCNN, RF, and SVM	Anthracnose, downy mildew, powdery mildew, and target leaf spots. Total images: 14208	93.4 using DCNN
[29]	Augmentation with image shifting, rotation, and mirroring	—	CNN	CNN	MYSV, ZYMV, CCYV, CMV, PRSV, WMV, KGMMV, and healthy. Total images: 7520	82.3
[30]	—	—	CNN	CNN	38821 leaves infected with any of 11 kinds of diseases. 1814 leaves infected with multiple diseases. 7676 healthy leaves	85.9 on multiple diseases 95.5 on entire dataset
Proposed method	Augmentation with flip vertically, horizontally, and rotation	—	CNN	CNN	Spider, leaf miner, downy mildew, powdery mildew, CYSDV, and healthy leaves	97.53 with unbalanced, 98.19 with balanced