Table 3.

Research status and problems of traditional machine learning methods.

Reference	Year	Purpose	Accuracy	Problems
[50]	2016	Combining HOG feature with Support Vector Machine (SVM) to identify grape leaves	83.50%	Single-feature detection has poor stability and low accuracy.
[35]	2016	Identifying different plant leaves on the basis of improved LBP	79.35%
[51]	2018	Using three shape features to compare the effect of SVM or Artificial Neural Network (ANN) on detecting sugar beets and weeds	93.33%	Analysis on the selection of features is lacking.
[52]	2009	Combining GW (Gabor wavelet) and GFD (gradient field distribution) to classify different weeds	93.75%
[53]	2015	Combining Gabor and Grey-level Co-occurrence Matrix (GLCM) to classify 31 plant leaves	91.60%	No actual field images are included, and the dataset is only composed of different plant leaves, without complex background, such as soil.
[54]	2017	Extracting the shape and texture features of an image to classify and recognize plant leaves	92.51%
[55]	2015	Using improved LBP and GLCM to categorize fresh tea in the production line	94.80%	Nonwhole plants are detected and recognized, and only the same kind of leaves is classified.