A plant image fine-grained classification method based on discriminant key domains and deep learning

Abstract

The invention relates to a plant image fine-grained classification method based on discriminant key domains and deep learning. According to the method, a CNN classification model considering a key domain and a global domain at the same time is used for carrying out fine-grained classification on to-be-classified images; the training process of the CNN classification model comprises the following steps: 1) obtaining an original plant image sample set, performing pixel-level semantic segmentation on an original plant image through DeepLab to obtain a discrimination key domain of a to-be-detectedtarget in the plant image, and forming a discrimination key domain image; 2) mixing the discrimination key domain image with an original plant image to form a classification training data set; and 3)training a GoogLeNet-based CNN classification model based on transfer learning, compared with the prior art, the method has the advantages of high classification accuracy, good robustness and the like, and the problem of low accuracy in fine-grained classification of plant images is solved.

Description

technical field [0001] The invention relates to an image fine-grained classification method, in particular to a plant image fine-grained classification method based on discriminant key domains and deep learning. Background technique [0002] At present, the classification of plant species mainly relies on the manual discrimination of experts in the field of plants, resulting in a large workload and low efficiency for image-based plant recognition tasks. After the rise of deep learning algorithms in the image field, automatic classification based on computer vision has been greatly developed. The literature "PANDA: Pose Aligned Networks for DeepAttribute Modeling" (Zhang N, Paluri M, Ranzato M, et al..2014:1637-1644) first uses the histogram of gradient orientation (HOG) and the component-based detection algorithm DPM and the Poselet method For the detection of bird targets and their local areas, CNN features are extracted for classification of the detection targets. The li...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a method for fine-grained classification of plant images based on discriminative key domains and deep learning in order to overcome the problem of low accuracy in the fine-grained classification of plant images in the prior art

Images