Plant leaf recognition system based on jigsaw disordered data disturbance mechanism

Abstract

The invention discloses a plant leaf recognition system based on a puzzle disruption type data perturbation mechanism. The system comprises a puzzle disruption module, an adversarial learning module, a modeling module, a loss function calculation and analysis module, and a cosine annealing learning module. The system processes an input sample into a random disruption form, forces a model to extract sub-region features, and carries out the recognition of the sub-region features; meanwhile, interference of a noise mode caused by a puzzle mechanism is avoided through adversarial learning, the situation that the model excessively fits interval information contained in puzzle input is avoided, meanwhile, correlation information between subareas is modeled through building learning, the space sensing ability of the model to the subareas is enhanced, and the model is more accurate. And finally, giving an opportunity of multiple error correction to the model through a cosine annealing scheme so as to improve the model recognition accuracy.

Description

technical field [0001] The invention relates to the technical field of plant leaf identification, in particular to a plant leaf identification system based on a puzzle-scrambling data disturbance mechanism. Background technique [0002] Identifying and classifying plants occupies a very important position in agricultural research. Correctly understanding and identifying plants can improve the pharmaceutical industry, ecosystem balance, and effectively increase agricultural productivity. The current mainstream traditional plant image classification work is generally done manually. Intensity has a high accuracy rate, but manual classification has the problem that with the increase of workload, the accuracy rate of workers will decrease, and the classification of plant image recognition technology can make up for this very well. Plant image recognition Technology generally refers to using the current mainstream computer vision technology to extract the features of the input ima...

AI Technical Summary

Problems solved by technology

[0002] Identifying and classifying plants occupies a very important position in agricultural research. Correctly understanding and identifying plants can improve the pharmaceutical industry, ecosystem balance, and effectively increase agricultural productivity. The current mainstream traditional plant image classification work is generally done manually. Intensity has a high accuracy rate, but manual classification has the problem that with the increase of workload, the accuracy rate of workers will decrease, and the classification of plant image recognition technology can make up for this very well. Plant image recognition Technology generally refers to using the current mainstream computer vision technology to extract the features of the input image, perform prediction classification, perform model training and structural adjustment on the data set, and finally realize the prediction of the image plant. Therefore, classification through plant image recognition technology has Great application advantages and prospects

[0003] In recent years, the field of computer vision has developed rapidly, which has brought great help to the recognition and classification tasks of many practical scenes. However, in complex plant leaf recognition scenes, because the background easily contains other types of samples or objects, and there are often intra-class The problem of too large spacing or too small spacing between classes, and the appearance of many plant leaves are only very small differences, so it will cause errors in the classification of recognition samples in the recognition system, or there are many samples with high similarity, making the parameters The increasing amount of data not only brings huge obstacles to the deployment and landing of the model, but also increases the difficulty of recognition, resulting in unsatisfactory recognition accuracy.

Images