Automatic crop disease and pest identification method suitable for fields

Abstract

The invention relates to the field of computer vision, in particular to an automatic crop disease and pest identification method suitable for fields. The method comprises the following steps: S1, acquiring crop image original data, and preprocessing the crop image original data; s2, inputting the preprocessed crop image original data into an improved crop disease and insect pest automatic identification model, and predicting a disease and insect pest category corresponding to the crop image original data, wherein the network architecture of the improved crop pest and disease damage automatic identification model is as follows: two branches of channel orthogonal constraint and species classification constraint are added to a backbone network of a convolutional neural network, and the channel orthogonal constraint is added to the last layer of features output by the backbone network, and the species classification constraints are added to any feature output by the backbone network. The method can achieve the accurate recognition of the types of diseases and pests, does not need a manager to have professional knowledge of domain experts, and improves the recognition performance of themodel in a field environment.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a field-adapted automatic identification method for crop diseases and insect pests. Background technique [0002] From 2006 to 2015, my country's crop diseases, insect pests, weeds and rodents were generally in a serious state, and the average annual loss of grain accounted for 20.88% of the country's total grain output. The sources of crop diseases and insect pests mainly include bacteria, fungi, oomycetes, viruses, nematodes and insects, etc. After the crops are infected with diseases, the leaves generally have symptoms such as spots, discoloration, deformity, wilting and necrosis. The health of crops is a condition for the survival of agricultural workers, and diagnosing these symptoms requires a high level of expertise, so it is of great significance to develop a method that can automatically identify crop diseases. [0003] Compared with traditional expert diagnosis methods t...

AI Technical Summary

Problems solved by technology

As the closest prior art of the present invention, the paper "Using Deep Learning for Image-Based Plant Disease Detection" and the paper "Solving Current Limitations of Deep Learning Based Approaches for Plant Disease Detection" have given detailed introductions, and the method has trained It is used to realize the classification neural network of plant diseases and insect pests, and realize the automatic identification of crop diseases and insect pests. They have high recognition accuracy under controlled laboratory conditions (with high requirements for lighting posture and background), but when it is applied The performance will drop sharply in the real field,

[0004] The existing methods for automatic identification of plant diseases and insect pests still have great limitations when solving the problem of automatic identification of field crop diseases and insect pests. The main problems are as follows: 1. The images of crops are usually collected under controlled conditions in the laboratory. In terms of crop growth conditions, crop images for identifying pests and diseases are too ideal, so the trained network does not perform well in actual measurements, and can only recognize single, frontal, uniformly illuminated, and single-background crop leaves

The above situation will reduce the accuracy of identifying crop pests and diseases

[0005] Considering the above limitations of the existing methods, if a more robust automatic identification algorithm for crop diseases and insect pests in the real field is proposed, the following three main problems will be encountered: 1. The background of the crop images in the data set is single, but The background of the leaf image in the wild environment is changeable, such as figure 1 as shown in (a)

Images