Convolutional-neural-network-based primary classification method for ear image of corn seed production

Abstract

The invention provides a convolutional-neural-network-based primary classification method for an ear image of corn seed production. The method comprises: an original training sample set and a testingsample set of corn ear images; on the basis of a migration learning method, discrimination is carried out by using an AlexNet convolutional neural network to obtain the type of each ear testing sampleand an actual type of each testing sample, and the accuracy rate of each determination result is determined; if the accuracy rate is in a preset range, the AlexNet convolutional neural network is optimized according to an expanded ear training sample set to obtain a second convolutional neural network, and the type of each testing sample in the testing sample set is determined again. According tothe invention, the tedious and one-sided problems of manual extraction of ear image features are solved by learning the ear image autonomously from low-level features like the color and side to high-level features like the angular point and the shape from a hidden layer and thus the convolutional neural network has capabilities of selecting image features autonomously and carrying out learning and identification. A novel method is provided for primary automatic ear selection of the corn seed production.

Description

technical field [0001] The invention relates to the field of pattern recognition, and more specifically, to a method for primary classification of ear images of corn seed production based on a convolutional neural network. Background technique [0002] In the process of corn seed production, seed companies usually perform ear selection first, and then perform grain selection and finishing. During ear selection, the normal corn ears with good phenotype characteristics are selected, and the corn ears with bracts are reclaimed for another treatment of removing bract leaves, and the corn ears of moth-eaten, mildew, grain abortion and heterogeneous corn are discarded. The quality and purity are closely related to increasing corn yield. Other corn ears include worm-eaten ears, moldy ears, heterogeneous ears, grain abortion and other ears, which are reflected in phenotypic characteristics. The shape, size, grain type, grain color, cob color, texture, etc. of corn ears are closely ...

AI Technical Summary

Problems solved by technology

The traditional method mainly relies on manpower to screen corn ears with bracts and other corn ears, which has the problems of slow screening speed and a large amount of manpower and financial resources

[0003] At present, computer vision technology is used to extract image features, identify and classify corn ears, instead of traditional manual selection, and use image processing technology to carry out fine digital analysis and processing of ear images, although it avoids the disadvantage of manpower and material resources. However, this method requires manual feature extraction and design, and the calculation is complicated. In addition, the actual seed production and processing factories in China are mostly carried out outdoors. assembly line

Images