A Crop Leaf Disease Image Recognition Method Based on Bilinear Residual Networks

Table 1

Summary of the related works.

Kind of model	Reference	Crop	Dataset	Accuracy	Advantage	Limitation
Global model	[11]	Tomato	PlantVillage	95.5%
	[12]	5 crops	Own dataset	96.3%
	[13]	10 crops	PlantVillage	85.22%
	[15]	Rice	Own dataset	95.48%
	[16]	Apple	Own dataset	78.8%
	[19]	Apple	Own dataset	93.71%
	[18]	Cassava	Kaggle cassava mosaic illness dataset	96.75%	Simple preprocessing, end-to-end deploying, low overhead, and easiness to use	Hard to extract fine-grained disease spots features accurately and completely
	[20]	Millet	Own dataset	95%
	[17]	Millet	Own dataset	98.78%
	[21]	Maize	PlantVillage and google websites combined	98.8%
Local model	[22]	2 crops	Own dataset	90%	Extract fine-grained disease spots features more accurately and completely	Additional operation and overhead, hard to be deployed in an end-to-end way
	[23]	Tomato	Own dataset	92.39%
	[24]	Tomato	Own dataset	96%
	[25]	Apple	PlantVillage	96.6%
	[26]	Guava	Own dataset	99%