Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Improved convolutional neural network-based plant leaf segmentation method

A convolutional neural network and plant leaf technology, applied in the field of image segmentation, can solve problems such as poor results, difficulty in extracting foreground objects, and mis-segmentation, so as to avoid under-segmentation and over-segmentation, overcome uneven illumination, and model parameters Small amount of effect

Inactive Publication Date: 2018-09-21
JIANGSU UNIV
View PDF2 Cites 15 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

As a typical threshold segmentation method, Otsu can automatically select the optimal segmentation threshold, which is relatively simple to implement, but the effect is not good when the foreground gray distribution range is large.
The clustering algorithm mainly clusters similar colors by calculating the similarity of colors between regions, but when the difference between the foreground and background colors is small, it is difficult to extract the foreground target
Due to the greater influence of external environmental factors, some existing methods have mis-segmentation phenomena, which will have an adverse effect on the subsequent recognition process, so it is necessary to propose an automatic and accurate segmentation method

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Improved convolutional neural network-based plant leaf segmentation method
  • Improved convolutional neural network-based plant leaf segmentation method
  • Improved convolutional neural network-based plant leaf segmentation method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0032] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0033] A kind of plant leaf segmentation method based on the improved fully convolutional neural network proposed by the present invention comprises the following steps:

[0034] (1) Test data acquisition and processing

[0035] (2) Build a fully convolutional neural network

[0036] (3) Training fully convolutional neural network

[0037] (4) Use the trained fully convolutional neural network model to achieve leaf segmentation

[0038] Described process (1) specifically comprises the following steps:

[0039] When selecting test samples, consider the complexity and diversity of the samples, such as inconsistencies in the size of the pictures in the picture collection, uneven lighting intensity, black shadows in the pictures, etc., and select diseased leaves with some diseased spots that are similar to the background color.

[0040] A total...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses an improved convolutional neural network-based plant leaf segmentation method. Eight-time up-sampling of original FCN is large in model parameter amount; in order to realize rapid and accurate segmentation of leaves, a direct connected structure is adopted to remove a part of the layers, single layer features are utilized to up-sample feature maps shrunk by a VGG16 model toan original map size according to a shrinking multiple through deconvolution, and three improved models are obtained. At last, 1762 pictures of 6 different plants are taken as training samples, 441 leave pictures are taken as test sample training models. Memory parameter occupation of direct connection type structure models which adopt four-time up-sampling is 6.90 MB, the shrinkage is 1 / 77 timeof the original FCN model, and the average correctness and average area overlap ratio of categories on test sets can be up to 99.099% and 98.204%. Compared with the tradition K-means cluster method and Otsu threshold value segmentation method, the method has the advantages that the average area overlap ratio is 3.704% and 4.295% higher, complete leaves can be extracted, and the influences of leavesurface color and illumination intensity unevenness are small.

Description

technical field [0001] The invention relates to an image segmentation method, in particular to a plant leaf segmentation method based on an improved fully convolutional neural network. Background technique [0002] Plants are an important part of the ecosystem, and plant identification and identification of plant diseases are important tasks in plant protection. Plant leaf segmentation aims to locate and extract the leaf area in the image to reduce the interference of background objects, which is an important step in plant identification and plant disease identification. The segmentation effect of leaves will have a direct impact on the subsequent feature extraction and pattern recognition work, so the problem of accurate segmentation and extraction of leaves has received extensive attention. [0003] For the segmentation of plant leaves, the existing methods are mainly divided into threshold-based segmentation and cluster-based segmentation. Threshold segmentation includes...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06T7/11G06N3/04G06N3/08
CPCG06N3/08G06T7/11G06T2207/30188G06T2207/20081G06N3/045
Inventor 孙俊谭文军武小红沈继锋陆虎戴春霞
Owner JIANGSU UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products