Apparatus and method for multimodal sensing and monitoring of perishable commodities

Patent metadata

Field	Value
Publication number	US-12038427-B2
Application number	US-201917296967-A
Country	US
Kind code	B2
Filing date	Oct 25, 2019
Priority date	Oct 29, 2018
Publication date	Jul 16, 2024
Grant date	Jul 16, 2024

Abstract

Health of perishable commodities such as eatables deteriorate over time. State of art systems for health monitoring of perishable commodities rely on measurement of limited parameters and also fail to consider effect of environment on the health of the perishable commodities. Disclosed herein is an apparatus and method for multimodal sensing and monitoring of perishable commodities. The apparatus allows to change environment within a closed chamber in which the perishable commodity being monitored is kept, and in turn allows to generate health data in different environment settings. This data is used to generate a health model. Data collected in real-time are processed with the health model to establish a correlation with at least one image, wherein each of such images in the health model represents certain health state. Based on the established correlation, health of the perishable commodity is determined.

First claim

What is claimed is:

1. An apparatus for monitoring health of a perishable commodity, comprising:
a housing with closed walls;
a sensor suite comprising a plurality of sensors, wherein the sensor suite is attached internally to the housing;
a support structure for keeping the perishable commodity within the housing, wherein the perishable commodity is kept in proximity of the plurality of sensors;
an environment controller for setting environment within the housing; and
a circuitry configured within the housing and communicably coupled with the sensor suite to determine the health of the perishable commodity, the circuitry comprising:
one or more hardware processors;
one or more communication interfaces; and
one or more memories storing a plurality of instructions, wherein the plurality of instructions when executed cause the one or more hardware processors to:
collect information pertaining to a distinct environmental setting from a user;
set the environment within a closed chamber to the distinct environmental setting by varying value of one or more environmental parameters;
collect value of health parameters corresponding to the perishable commodity being monitored automatically based on inputs from a plurality of sensors and the plurality of sensors are selected automatically based on a type of perishable commodity being monitored and the health parameters to be measured, corresponding sensors for varied type of perishable commodities are pre-configured or dynamically configured in the memory;
establish a correlation between the collected value of the health parameters with at least one time series image representing a health state, by comparing the collected value of the health parameters including temperature, humidity, amount of ethylene, CO2 rate, ethylene rate, freshness index (FI) with a pre-trained health model, the pre-trained health model trained on sensor and image data indicating change in health of a plurality of perishable commodities with time under plurality of environmental settings, wherein the health model is trained on a plurality of features extracted from a plurality of time series sensory data collected from the plurality of perishable commodities, wherein the time series sensory data is annotated as representing a degree of freshness level, wherein the health model is unique to each type of perishable commodity being monitored, and generated using values of the health parameters, wherein the freshness index includes a set of values, each value representing a particular health condition, value 1 of the freshness index indicate that the perishable commodity being monitored is fresh while value 10 indicate that the perishable commodity being monitored is degrading,
wherein the degree of freshness level indicates varied stages with respect to freshness of the perishable commodity, wherein the varied stages include not ripen, medium ripen, fully ripen followed by determining a Ripening Index (RI) with values indicating minimum ripening and maximum ripening, and
determine health of the perishable commodity based on the established correlation and determine amount of time left for the perishable commodity to reach a particular degree of freshness.

2. The apparatus as claimed in claim 1 , wherein the health model is a combination of a plurality of machine learning algorithms.

3. The apparatus as claimed in claim 2 , wherein the plurality of machine learning models comprise at least one Convolution Neural network (CNN) model, wherein the at least one CNN model is trained with a plurality of time series images to extract a plurality of vector representations which represent degradation in a plurality of perishable commodities, further wherein the vector representation of the time series images are annotated as representing a degree of freshness level.

4. The apparatus as claimed in claim 2 , wherein the plurality of machine learning models comprise at least one Support Vector Regression (SVR) model.

5. A processor implemented method ( 300 ) for monitoring health of a perishable commodity using an apparatus ( 100 ), comprising:
collecting information pertaining to a distinct environmental setting from a user, via one or more hardware processors;
setting the environment within a closed chamber to the distinct environmental setting by varying value of the one or more environmental parameters, via the one or more hardware processors;
collecting value of health parameters corresponding to the perishable commodity being monitored automatically based on inputs, from a plurality of sensors and the plurality of sensors are selected automatically based on a type of perishable commodity being monitored and the health parameters to be measured, corresponding sensors for varied type of perishable commodities are pre-configured or dynamically configured in a memory, via the one or more hardware processors;
establishing a correlation between the collected value of the health parameters with at least one time series image representing a health state, by comparing the collected value of health parameters including temperature, humidity, amount of ethylene, CO2 rate, ethylene rate, freshness index (FI) with a pre-trained health model, the pre-trained health model trained on sensor and image data indicating change in health of a plurality of perishable commodities with time under plurality of environmental settings, wherein the health model is trained on a plurality of features extracted from a plurality of time series sensory data collected from the plurality of perishable commodities, wherein the time series sensory data is annotated as representing a degree of freshness level, via the one or more hardware processors, wherein the health model is unique to each type of perishable commodity being monitored, and generated using values of the health parameters, wherein the freshness index includes a set of values, each value representing a particular health condition, value 1 of the freshness index indicate that the perishable commodity being monitored is fresh while value 10 indicate that the perishable commodity being monitored is degrading,
wherein the degree of freshness level indicates varied stages with respect to freshness of the perishable commodity, wherein the varied stages include not ripen, medium ripen, fully ripen followed by determining a Ripening Index (RI) with values indicating minimum ripening and maximum ripening, and
determining health of the perishable commodity based on the established correlation and determine amount of time left for the perishable commodity to reach a particular degree of freshness.

6. The method as claimed in claim 5 , wherein the health model is a combination of a plurality of machine learning algorithms.

7. The method as claimed in claim 6 , wherein the wherein the plurality of machine learning models comprise at least one Convolution Neural network (CNN) model, wherein the at least one CNN model is trained with a plurality of time series images to extract a plurality of vector representations which represent degradation in a plurality of perishable commodities, further wherein the vector representation of the time series images are annotated as representing a degree of freshness level.

8. The method as claimed in claim 6 , wherein the plurality of machine learning models comprise at least one Support Vector Regression (SVR) model.

9. One or more non-transitory machine readable mediums comprising one or more instructions which when executed by one or more hardware processors for monitoring health of a perishable commodity using an apparatus ( 100 ), causes:
collecting information pertaining to a distinct environmental setting from a user;
setting the environment within a closed chamber to

Assignees

• Tata Consultancy Services Ltd

Inventors

• Dutta Jayita
• Deshpande Parijat
• Rai Beena

Classifications

• G06T2207/30128

  Food products · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T7/0004

  Industrial image inspection · CPC title

• G06Q10/087

  Inventory or stock management, e.g. order filling, procurement or balancing against orders · CPC title