# US 7,610,155 B2 · Methods for processing spectral data for enhanced embryo classification > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 7,610,155 B2 - Document ID: US7610155B2 - Status: Granted - Kind: B2 - Application number: 11/297,919 - Filed: 2005-12-08 - Published: 2006-06-29 - Granted: 2009-10-27 - Earliest claimed: 2004-12-28 - Assignees: Weyerhaeuser NR Company - Inventors: Roger Timmis; Mitchell R. Toland - Web page: https://patentgrep.com/patent/US7610155B2 ## CPC - G06V 20/695: {Preprocessing, e.g. image segmentation} - G06V 20/698: {Matching; Classification} - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract A method is disclosed for classifying plant embryos according to their quality using a logistic regression model. First, sets of image or spectral data are acquired from plant embryos of known quality, respectively. Second, each of the acquired sets of image or spectral data is associated with one of multiple class labels according to the corresponding embryo's known quality. Third the sets of image or spectral data values are filtered to provide filtered image or spectral data values. Fourth, a classification algorithm, e.g., a logistic regression analysis is applied to the filtered data values and their corresponding class labels to develop a classification model. Fifth, image or spectral data are acquired from a plant embryo of unknown quality, and filtered data values are derived therefrom. Sixth, the classification model is applied to the filtered data values for the plant embryo of unknown quality to classify the same. ## Claims ### Claim 1 A method for developing a classification model for classifying plant embryos according to their quality, the method comprising: (a) receiving sets of spectral data of plant embryos of known quality, each set of spectral data of a plant embryo of known quality being associated with one of multiple class labels according to the known quality of the corresponding embryo from which the set of spectral data of the plant embryo of known quality is obtained; (b) filtering, using a plurality of filters, the received sets of spectral data of the plant embryos of known quality to provide sets of filtered spectral data values of the plant embryos of known quality, each set of filtered spectral data values of the plant embryo of known quality being associated with one of the multiple class labels according to the known quality of the corresponding embryo from which the set of filtered spectral data values of the plant embryo of known quality values is derived, wherein the filtering comprises: (i) for each filter, linearly combining spectral data values over a band of wavelengths to provide linearly combined spectral data values; (ii) identifying those filters that best discriminate between different embryo class labels based on cumulative distribution functions estimated at each wavelength for each filter for each embryo class label, to provide sets of best-filtered spectral data values; and (iii) reducing the sets of best-filtered spectral data values based on the Smirnov statistics calculated between pairs of embryo class labels at each wavelength for each filter, to provide the sets of filtered spectral data values; (c) applying a classification algorithm to the sets of filtered spectral data values of the plant embryos of known quality and their corresponding class labels to develop the classification model; and (d) outputting the developed classification model to a user for use in classifying embryos according to their quality, wherein all of the steps are performed on a computer. ### Claim 2 (depends on 1) The method of claim 1, further comprising, between step (a) and step (b), the step of pre-processing the received sets of spectral data of the plant embryos of known quality before the filtering step. ### Claim 3 (depends on 1) The method of claim 1, wherein the classification algorithm comprises a logistic regression. ### Claim 4 (depends on 1) The method of claim 1, wherein there are three or more class labels. ### Claim 5 (depends on 1) The method of claim 1, wherein the plant is a tree. ### Claim 6 (depends on 5) The method of claim 5, wherein the tree is a member of the order Coniferales. ### Claim 7 (depends on 6) The method of claim 6, wherein the tree is a member of the family Pinaceae. ### Claim 8 (depends on 1) The method of claim 1, wherein the plant embryos are plant somatic embryos. ### Claim 9 (depends on 1) The method of claim 1, further comprising, after step (d), the steps of: (e) acquiring spectral data of a plant embryo of unknown quality; (f) filtering the spectral data of the plant embryo of unknown quality using one or more of a plurality of filters; and (g) applying the classification model to the filtered spectral data values of the plant embryo of unknown quality to classify the same. ### Claim 10 An article comprising a computer-readable storage medium including computer-executable instructions, wherein the instructions when loaded onto a computer perform the steps comprising: (a) receiving sets of filtered spectral data of plant embryos of known quality, each set of filtered spectral data of a plant embryo of known quality being associated with one of multiple class labels according to the known quality of the corresponding embryo from which the set of filtered spectral data of the plant embryo of known quality is obtained, wherein the sets of filtered spectral data are provided based on filtering using a plurality of filters comprising the steps including: (i) for each filter, linearly combining spectral data over a band of wavelengths to provide linearly combined spectral data; (ii) identifying those filters that best discriminate between different embryo class labels based on cumulative distribution functions estimated at each wavelength for each filter for each embryo class label, to provide sets of best-filtered spectral data; and (iii) reducing the sets of best-filtered spectral data based on the Smirnov statistics calculated between pairs of embryo class labels at each wavelength for each filter, to provide the sets of filtered spectral data; (b) applying a classification algorithm to the sets of filtered spectral data of embryos of known quality and their corresponding class labels to develop a classification model; and (c) outputting the developed classification model to a user for use in classifying embryos according to their quality. ### Claim 11 (depends on 10) The article of claim 10, wherein the instructions further perform the steps of selecting a portion of the filtered spectral data of the plant embryos of known quality and applying the classification algorithm to the portion of filtered spectral data of embryos of known quality and their corresponding class labels to develop the classification model. ### Claim 12 (depends on 10) The article of claim 10, wherein the instructions further perform the step of applying the classification model to filtered spectral data of a plant embryo of unknown quality to classify the same. ### Claim 13 (depends on 10) The article of claim 10, wherein the instructions further perform the step of filtering spectral data of a plant embryo of unknown quality to provide the filtered image spectral data of the plant embryo of unknown quality.