Apple virus identification method based on deep learning

Abstract

The invention discloses an apple virus recognition method based on deep learning, and the method comprises the steps: building a fruit damage recognition network through the improvement of an originalresidual network structure and a loss function based on deep learning; for a training problem under a small sample, carrying out model training by combining strategies of transfer learning and a hierarchical learning rate in a network training process; and compressing the finally obtained model, so as to reduce the deployment cost of the model and improve the recognition efficiency. Theoretical guidance and technical support are provided for crop disease recognition.

Description

technical field [0001] The invention relates to the field of apple virus identification. Background technique [0002] Crop virus recognition plays an important role in crop growth. There are many methods for virus identification. Although the traditional image technology does not require a high number of samples, it consumes a lot of human resources and is easily affected by empiricism; the hyperspectral image method can judge from the perspective of spectral distribution. crops, but its high cost and strong equipment dependence make it difficult to popularize and use; and the method of deep learning not only overcomes the shortcomings of manual feature extraction, but also can be easily extended to orchards, so the deep learning method is adopted method for apple disease detection. Contents of the invention [0003] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides an apple virus identification method based...

AI Technical Summary

Problems solved by technology

[0006] (1) There are many neural network structure parameters in the existing apple virus identification algorithm, which is easy to cause over-fitting when used to train apple disease classification. An improved neural network structure of the residual network is proposed. By optimizing the original residual The network convolution kernel is composed to reduce the number of parameters; and for the problem that the characteristics of different diseases are similar and easy to be misidentified, a penalty item for similarity between classes is added to the traditional loss function to improve the accuracy of disease recognition;

Images