A crop image segmentation system and method based on deep neural network modeling

Abstract

The invention discloses an image segmentation system based on deep neural network modeling. The system comprises an image acquisition module, a pixel classification module configured to manually obtain two types of pixels in the crop image in a clicking manner, namely, crop pixels and background pixels, namely, positive samples and negative samples corresponding to class labels respectively, and select the same number of positive samples and negative samples as training samples of a deep convolution neural network; a color space conversion module configured to convert the training sample froman RGB color space to a standardized rgb and Lab color space, and convert the Lab color value of the sample into an unsigned 8-bit integer form according to an ICC specification to form a color feature of the training sample; a neural network training module; and a model test module. The system of the invention has the advantages of higher crop image segmentation processing speed and segmentationaccuracy, can be better adapted to the outdoor complex and changeable illumination environment, and can effectively segment and extract crops in crop growth observation.

Description

technical field [0001] The present invention relates to a crop image segmentation system and method, in particular to an image segmentation system and method based on deep neural network modeling. Background technique [0002] Since rice is a crop that is easily affected by disastrous climatic conditions and pests and diseases, the above disasters will directly affect the yield of rice. Therefore, it is very necessary to observe the growth status of rice, which is conducive to the timely assessment of the disaster and the timely development of countermeasures when rice disasters occur. In addition, reasonable field operations such as irrigation, fertilization, and sun-drying based on rice observation results during the general rice development process can also effectively increase rice production. The traditional observation method is mainly manual observation, that is, some agricultural technicians judge the growth status of rice based on their personal observation experie...

AI Technical Summary

Problems solved by technology

[0004] However, in the above methods, whether ExG, ExGExR or CIVE algorithms are threshold-based crop segmentation methods, the advantage of this type of method is that the segmentation speed is fast and real-time processing can be achieved, but in complex lighting environments, threshold-based crop segmentation methods often Does not bring stable segmentation results

And these three types of methods are not applicable to the segmentation of non-green crops such as purple cabbage, etc.

The EASA segmentation method based on Bayesian theory generally requires relatively complete training samples to ensure the unbiasedness of the training data, so that the method requires a large sample size to ensure the completeness of the training sample set. The data collection process of this method is relatively cumbersome

The advantage of the AP-HI method is that it can realize the modeling of the crop color model through a small number of samples. However, the defect of this method is that not all the color distributions of the crops satisfy the strong assumption of a single Gaussian distribution.

Images