Plant phenotype detection system and method based on attention and multiple knowledge migration

Abstract

The invention discloses a plant phenotype detection system and method based on attention and multiple knowledge migration, and relates to the field of plant phenotype intelligent identification. The system comprises an industrial camera (10), a server (20) and embedded equipment (30) which are communicated in sequence, a data set making module (21), a teacher target detection model (22) and a student target detection model (23) which interact in sequence are embedded in the server (20); and a real-time acquisition module (31), a final model (32) and an output module (33) which interact in sequence are embedded in the embedded equipment (30). According to the invention, a mixed domain attention module and a corresponding attention loss function are improved; a feature fusion module and a corresponding feature fusion layer knowledge migration loss function are designed; a knowledge migration training method based on multiple losses is provided; real-time detection of the plant phenotypein the natural environment is achieved, and the method is suitable for plant phenotype research and has wide prospects.

Description

technical field [0001] The invention relates to the field of plant phenotype intelligent recognition, in particular to a plant phenotype detection system and method based on attention and multiple knowledge transfer. Background technique [0002] Research on plant phenotypes can not only provide theoretical basis and technical means for crop breeding, cultivation and agricultural production, but also contribute to the precision and sustainable development of agricultural production. Traditional plant phenotype research mainly relies on manual observation and measurement to obtain a description of the physical properties of plants. This method often relies on manual detection of individual traits from small sample plants, so the amount of data that can be obtained is very limited and the efficiency is very low. [0003] At present, most of the plant phenotype data are mainly image data, so with the advantage of deep learning in feature extraction, the identification of plant ...

AI Technical Summary

Problems solved by technology

For example, the paper "StalkNet: A deep learning pipeline for high-throughput measurement of plant stalk count and stalkwidth" published by Baweja et al. in Field and Service Robotics2018: 271-284 uses a ground robot equipped with a high-resolution stereo imager to capture sorghum plant experiments Based on the dense image data of the plot, the measurement algorithm of the number of stems and stem width is constructed through a fast regional convolutional neural network (Faster-Regions with CNN Features, Faster-RCNN). This method has high algorithm accuracy, but the detection speed Significantly slower than the single-stage target detection algorithm; Sarker and Kim published a paper "Farm land weed detection with region-based deep convolutional neural networks" on electronic bulletin board online 2019-06-05, proposing a region-based fully convolutional network to achieve For the identification of weeds in the complex environment of the field, although the actual test shows a high accuracy of weed identification, there is still a lot of room for improvement

The above-mentioned related methods have achieved relatively leading advantages in feature extraction compared with previous feature extraction methods, but it is often difficult to balance accuracy and real-time performance in the face of plant phenotype detection in complex backgrounds. Accuracy and real-time requirements cannot be ignored

Images