Method for realizing plant leaf scab segmentation and identification by multi-scale deconvolution network

Abstract

The invention provides a method for realizing plant leaf scab segmentation and identification by using a multi-scale deconvolution network, which realizes end-to-end plant leaf scab segmentation and identification by using a small amount of pixel-level marks. The method comprises the following steps: firstly, constructing a multi-scale feature extraction module by utilizing a multi-scale residualblock, and extracting multi-scale disease features; then, introducing a classification and bridging module to obtain an activation graph of a specific class, wherein the activation graph comprises keyinformation of disease spots of the specific class, and conducting up-sampling on the activation graph, so segmentation of the disease spots is achieved; and finally, designing a deconvolution module, extracting a real position of a network concerned disease spot by combining a small amount of disease spot annotation guide features, and further optimizing an identification and segmentation effect. The method provided by the invention can be suitable for identifying and segmenting plant leaf diseases with insufficient pixel-level labeling samples, and realizes the integration of identificationand segmentation. The model has high robustness in disease images with insufficient light and noise interference.

Description

technical field [0001] The invention belongs to the field of plant disease detection, in particular to a multi-scale deconvolution network to realize segmentation and identification of plant leaf disease spots. Background technique [0002] Crop diseases are one of the main reasons for the decline in the yield of agricultural products. Timely and effective analysis of the disease spot characteristics of crops will help to quickly judge the type and degree of crop diseases, provide corresponding guidance and suggestions for disease control, and reduce economic losses. At present, there are mainly two types of methods for the classification of plant diseases. One mainly relies on manual design feature extraction, and uses machine learning methods to classify features. This method generally needs to segment diseased spots or diseased leaves first, and the workload is relatively large, and feature extraction methods must be designed for different disease combinations each time...

AI Technical Summary

Problems solved by technology

[0004] The technical problem solved by the present invention is to provide a multi-scale deconvolution network to realize the segmentation and identification of plant leaf disease spots. Based on the labeling of disease categories and a small number of disease spots, the multi-scale feature extraction module is used to extract multi-scale disease features. Introduce the classification and bridging module to obtain the activation map of a specific class, use the deconvolution network to achieve lesion segmentation, and use a small number of lesion labels to guide the feature extraction network to focus on the real location of the lesion, further optimize the recognition and segmentation effect, and realize the recognition and segmentation. Integrated, the model has strong robustness in disease images with insufficient light and noise interference, and can be applied to the identification and segmentation of plant leaf diseases with insufficient number of disease spot labeling samples

Images