From: Plant diseases and pests detection based on deep learning: a review
Technology | Traditional image processing methods | Deep learning methods |
---|---|---|
Essence | Manual design features + classifiers (or rules) | Automatic learning of features from large amounts of data |
Method | Image segmentation method: Threshold segmentation; Roberts, Prewitt, Sobel, Laplace and Kirsh edge detection; region segmentation Feature extraction method: SIFT, HOG, LBP, shape, color and texture feature extraction method Classification method: SVM, BP, Bayesian | CNN |
Required conditions | Relatively harsh imaging environment requirements, high contrast between lesion and non-lesion areas, less noise | Adequate learning data and high-performance computing units |
Applicable scenarios | It is often necessary to change the threshold or redesign the algorithm when imaging environment or plant diseases and pests class changes, which has poor recognition effect in real complex natural environment | It has ability to cope with certain real and complex natural environment changes |