Grape leaf scab identification method based on fine-grained generative adversarial network

Abstract

The invention belongs to the field of artificial intelligence and plant protection, relates to interdisciplinary cross application of artificial intelligence and plant protection subjects, and particularly relates to a grape leaf scab identification method based on a fine-grained generative adversarial network. Comprising the steps of data collection and labeling, significant scab region detection and segmentation, fine-grained generative adversarial network image enhancement, training of a deep learning classification model, and grape leaf scab identification by using the trained model. The method mainly solves the problem that the leaf scab recognition rate is low under the conditions that grape leaf diseases are in the form of scab, in the early stage of morbidity, novel diseases, rare scab or insufficient training samples, is mainly used for scab recognition in the early stage of morbidity of the grape leaf scab, and can take corresponding intervention measures as soon as possible. A foundation is laid for precise pesticide application in the next step, economic losses are reduced to the maximum extent, the dosage can be reduced, and the environment is protected. The method can also be expanded to the situation that other plant leaf diseases are scabs.

Description

Technical field [0001] The present invention relates to artificial intelligence and plant protection related art, in particular, invented a fine grape blades of Fine grained-GaN, invented. Background technique [0002] The common diseases of grapes mainly have black rot, leaf blight, blacka ramp (Black Measles), etc., from the form of expression, these grapes are mainly manifested in countless The spatum composition, timely discovery of early characteristics of grape disease and conduct corresponding interventions to control the spread of grape diseases, to make corresponding interventions as soon as possible, laying the foundation for the next precision, minimize reduction Economic losses can also reduce the amount of medication and protect the environment. Due to early diseases, it is not easy to detect, especially for rare diseases, or new diseases, this most obvious feature is that training data is insufficient, lacking prior trained recognition model available. Solving this ...

AI Technical Summary

Problems solved by technology

[0010] (1) Lack of data sets required for grape leaf lesion detection

[0011] (2) Existing methods mainly focus on the research of common grape diseases that have occurred on a large scale, and there are few methods for identifying early diseases of grape leaves or identifying new / rare diseases

[0012] (3) If rare diseases or new diseases appear in the actual planting environment, it is very troublesome to collect data in this case. In the absence of sufficient training sets, the deep learning model often has the problem of overfitting, thus High recognition accuracy cannot be guaranteed

Images