Industrialized system for rice grain recognition and method thereof

Patent metadata

Field	Value
Publication number	US-2022012519-A1
Application number	US-202017440202-A
Country	US
Kind code	A1
Filing date	Mar 19, 2020
Priority date	Mar 19, 2019
Publication date	Jan 13, 2022
Grant date	—

Abstract

An industrialized system and method for rice grain recognition. An optical image is taken and transmitted to a digital platform, wherein the system segments the optical image and extracts and/or measures appropriate grain features from the image describing different aspects of the grain. The image is processed by the system which includes a selector selecting different machine learning structures, applying the different machine learning structures to the extracted features for rice grain recognition, and selecting the best of the applied machine learning structures by a random sampling process. The selected best of the applied machine learning structures is further optimized by varying an appropriate threshold by a threshold trigger based on a confusion matrix comprising. An active learning structure based on the confusion matrix to the user. The system is retrained based on the feedback parameters of the feedback loop.

First claim

1 - 17 . (canceled)

18 . An industrialized system for rice grain recognition, the system comprising:
an optical capturing device for capturing optical images, the optical capturing device comprising a data interface for transmitting the optical images to a digital platform over a data transmission network, wherein the optical images are analyzed at the digital platform and a response to the rice recognition is provided to the optical capturing device and/or a user device; and circuitry at the digital platform, the circuitry being configured to implement
a segmentation module segmenting the captured optical image for grains by identifying optical image segments capturing grains or grain parts; and
a feature extractor extracting measurable grain features from the identified grains or grain parts in the optical image, the features describing different parametrized aspects of the grains by a feature extraction process at least comprising shape parameter values and color parameter values and/or spatial parameters and/or geometric parameters,
wherein, in a learning mode, the circuitry is configured to implement a classifier with a selector for sequential selecting of a plurality of machine learning structures, the selector applying the different machine learning structures to the extracted grain features for rice grain recognition, and selecting an applied machine learning structure of the applied machine learning structures, which has a number of grains classified as false positive as low as possible while keeping the overall accuracy as high as possible, and wherein the selected machine learning structure of the applied machine learning structures is further optimized by varying a discrimination threshold by a threshold trigger, wherein the classified false positives are further minimized to a determined threshold probability for a grain to be classified as good, so that the variation of the discrimination threshold vanes the diagnostic ability of the binary classifier system related to the variation of the discrimination threshold.

19 . The industrialized system according to claim 18 , wherein the threshold trigger triggers an optimal threshold parameter value in a 2-dimensional optimization process measuring the true-positive rate against the false-positive rate at various threshold settings optimizing the classifier's sensitivity as a function of its fall-out.

20 . The industrialized system according to claim 19 , wherein the threshold parameter value is optimized by the threshold trigger in such a way that the false-positive rate is limited to a predefined trigger value.

21 . The industrialized system according to claim 20 , wherein the predefined trigger value for the false-positive rate is ≤1%.

22 . The industrialized system according to claim 18 , wherein the threshold trigger triggers an optimal threshold parameter value in a 2-dimensional optimization process measuring the true-positive rate against the false-positive rate at various threshold settings optimizing the classifier's sensitivity as a function of its fall-out by a confusion matrix.

23 . The industrialized system according to claim 18 , wherein an active learning structure based on a confusion matrix comprising the values of True Positive (TP), False Negative (FN), False Positive (FP) and True Negative (TN) for the classified rice grains, providing a feedback loop as an iterative retraining process to a user or human expert, wherein the system is retrained based on the feedback parameters of the feedback loop using the segmentation process and/or the feature extraction process and/or the classification process in the training mode.

24 . The industrialized system according to claim 18 , wherein the selectable machine learning structures comprise at least one neural network structure.

25 . The industrialized system according to claim 24 , wherein the at least one neural network structure comprises at least one Convolutional Neural Network (CNN).

26 . The industrialized system according to claim 18 , wherein the extracted grain features from the identified grains of the optical image comprise at least shape parameters and/or color parameters and/or spatial parameters and/or geometric parameters of the grains.

27 . The industrialized system according to claim 18 , wherein the optical capturing device is applied to a light tent comprising a uniform light source.

28 . The industrialized system according to claim 27 , wherein the optical capturing device is a mobile optical capturing device which is applied externally to the light tent via a camera hole in the light tent.

29 . The industrialized system according to claim 18 , wherein the optical capturing device is a mobile smart phone, and wherein the optical image is captured by a camera of the mobile smart phone using a dedicated mobile app and transmitted via one of the data transmission interfaces of the mobile smart phone over the data transmission network to the digital platform.

30 . The industrialized system according to claim 18 , wherein the circuitry comprises a split processing structure with a cloud computing platform providing the rice grain recognition as software as a service and a machine learning/data mining system providing the machine learning and data mining processing through a modular data pipelining structure.

31 . The industrialized system according to claim 18 , wherein, for the selection of the plurality of machine learning structures, the selector of the classifier comprises a sampling process based on a random sampling process, and wherein a machine learning structures is selected by the selector randomly, such that each machine learning structure has the same probability of being chosen during the sampling process.

32 . The industrialized system according to the claim 31 , wherein each of the selected machine learning structures is trained based on a random forest structure providing an ensemble learning process for the classification of the grains using the plurality of machine learning structures.

33 . The industrialized system according to the claim 32 , wherein each of the selected machine learning structures is trained by a random forest learner providing an appropriate random forest predictor wherein each random forest predictor is optimized by varying an appropriate threshold by the threshold trigger, and wherein the variation of the discrimination threshold varies the diagnostic ability of the binary classifier system varies related to the variation of the discrimination threshold.

34 . An industrialized method for a system for rice grain recognition, the method comprising:
capturing optical images, using an optical capturing device; transmitting the optical images to a digital platform over a data transmission network via a data interface; analyzing the optical images, using circuitry at the digital platform; providing a response to the rice recognition to the optical capturing device and/or a user device; and using a segmentation module implemented by the circuitry, automatically segmenting the captured and transmitted optical image for grains by identifying optical image segments capturing grains or grain parts, wherein, using a feature extractor implemented by the circuitry, measurable grain features are extracted from the identified grains or grain parts in the optical image, the features describing different parametrized aspects of the grains by a feature extraction process at least comprising shape parameter values and color parameter values and/or spatial parameters and/or geometric para

Assignees

• Buehler Ag

Inventors

• Treboux Jérôme
• Mayoraz Calixte
• Genoud Dominique
• Graeber Matthias
• Larsen Sara

Classifications

• G06V10/82

  using neural networks · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06F18/24323

  Tree-organised classifiers · CPC title

• G06V10/462 Primary

  Salient features, e.g. scale invariant feature transforms [SIFT] · CPC title

• G06F18/2178

  based on feedback of a supervisor · CPC title