Improved AlexNet with Inception-V4 for Plant Disease Diagnosis

Table 1

Summary of related works on plant disease classification.

Reference	Plant types	Dataset	Data augmentation	Methods	Limitation
Srdjan et al. [7]	13 kinds of plants	Stanford background dataset	Image transformations used for augmentation: (a)affine transformations; (b)perspective transformations; (c) rotations.	CNN	Training less data
Mohanty et al. [8]	14 crop diseases	PlantVillage	Resize the images to 256 × 256 pixels, and perform both the model optimization and predictions on these downscaled images	AlexNet	When tested on a set of images taken under conditions different from the train images, the accuracy is reduced substantially to just above 31%
Dyrmann et al. [13]	22 crop samples	BBCH12e16	—	DCNN	Due to the small number of training samples, the recognition accuracy fluctuates greatly
Ferreira et al. [14]	Soybean crops diseases	Captured by the UAV	—	ConvNets or CNNs	Dependency on feature extractors
Ghazi et al. [15]	1,000 species of trees, herbs, and ferns	LifeCLEF 2015	Decrease the chance of overfitting, image transforms such as rotation, translation, reflection, and scaling	GoogleNet, AlexNet, and VGGNet	As an example, increasing the batch size from 20 to 60 increases the training time 3-fold but does not match the performance obtained by increasing the number of iterations by the same amount
Liu et al. [16]	16 kinds of insect pests	Multi-class pest dataset 2018 (MPD2018)	—	CNN	The model did not do a good job of identifying similar pests in different categories methods
Geetharamani and Arun Pandian [9]	13 different of plant leaves	PlantVillage	Image flipping, gamma correction, noise injection, PCA color augmentation, rotation, and scaling transformations	Deep CNN	The model can only identify leaf diseases, but it cannot identify other parts of the plant diseases
Ozguven and Adem [6]	Sugar beet leaf disease	Sugar beet leaf images dataset	—	Faster R–CNN	The accuracy of disease detection is low
Chao et al. [17]	Apple tree leaf diseases	Laboratory independent planting and cultivation	Image scaling, dataset expansion, and dataset normalization	DCNN	There are few types of data sets, and the specific network architecture of various structures lacks a description