Pattern recognition apparatus and pattern recognition method

A pattern recognition apparatus comprising:

• pattern recognition means being set to an initial state before a plurality of patterns are inputted, for classifying said patterns to classification classes to which said patterns belong, evaluating reliability of the classification result and outputting the classification classes to which said patterns belong, as recognizable for those of said patterns which provide the classification result having high reliability;
• first storage means for storing those of said patterns for which said classification results are evaluated as having low reliability and recognition of which is suspended;
• second storage means for storing those of said patterns for which said classification results are evaluated as having high reliability by said pattern recognition means and said classification classes outputted from said pattern recognition means;
• pattern recognition method adjustment means for optimizing said pattern recognition means by using said patterns and said classification classes stored in said second storage means as a training sample;
• and
• reference pattern recognition method storage means for storing the initial state of said pattern recognition means;
• wherein:
• after said pattern recognition means is optimized by said pattern recognition method adjustment means, said patterns stored in said first storage means are inputted to said pattern recognition means and said classification classes to which said patterns belong are outputted.

A pattern recognition apparatus comprising:

• pattern storage means for storing a plurality of patterns;
• a neural network for executing pattern recognition;
• reliability evaluation means for evaluating reliability of the pattern recognition result of said patterns by the output value of said neural network;
• initial weight value storage means for storing the number of layers of said neural network, the number of neurons of each layer and a weight value of each neuron under the initial state of said neural network;
• neural network training means for training said neural network;
• and
• pattern recognition result storage means for storing the pattern recognition result;
• wherein:
• after a plurality of said patterns are stored in said pattern storage means and said neural network is initialized in accordance with the content of said initial weight value storage means;
• (a) a process for executing pattern recognition by inputting said patterns stored in said pattern storage means to said neural network;
• (b) a process for storing said pattern recognition result in said pattern recognition result storage means;
• (c) a process for evaluating reliability of said pattern recognition result of said patterns by using the output value of said neural network by said reliability evaluation means;
• and
• (d) a process for executing training of said neural network for said patterns evaluated as having high reliability by said neural network training means by using said pattern recognition result of said patterns as a desired output;
• are executed for all of said patterns stored in said pattern storage means, and after the processes from said item (a) to said item (d) are repeated until reliability of all of said pattern recognition result of said patterns is evaluated as being high by said neural network or until a predetermined number of times is reached, said pattern recognition result stored in said pattern recognition result storage means is outputted.

A pattern recognition apparatus comprising:

• pattern recognition means for classifying a plurality of inputted pattern to classification classes to which said patterns belong, evaluating reliability of the classification result, outputting said classification result as recognizable for those of said patterns which provides said classification result having high reliability, suspending recognition for those of said patterns for which said classification results have low reliability, and outputting a first applicant and a second applicant of said classification class to which said patterns are to be classified;
• storage means for storing said first applicant and said second applicant;
• a counter for counting the number of each kind of objects for which said classification class is outputted by said pattern recognition means;
• and
• reasoning means for deciding said classification class of said patterns by using the content of said counter and said first applicant and said second applicant stored in said storage means.

A pattern recognition method comprising:

• a first step for executing first pattern recognition by classifying a plurality of patterns to classification classes to which said pattern belong, evaluating reliability of the classification result, and obtaining the classification classes to which said patterns belong, as recognizable, for those of said patterns which provide said classification result having high reliability;
• a second step for storing those of said patterns for which said classification results are evaluated as having low reliability and recognition of which is suspended by said first step in first storage means;
• a third step for storing said patterns for which said classification results are evaluated as having high reliability and said classification classes by said first step in second storage means;
• a fourth step for obtaining said classification classes to which said pattern belong, from said patterns stored in said first storage means by using said patterns and said classification classes stored in said second storage means as a training sample;
• and
• a fifth step for constructing said second pattern recognition by using said patterns and said classification classes stored in said second storage means as a training sample.

A pattern recognition method comprising:

• a first step being set to an initial state before a plurality of patterns are inputted, for classifying said patterns to classification classes to which said patterns belong, evaluating the classification result and obtaining said classification classes to which said patterns belong, for those of said patterns for which the classification result having high reliability is obtained, as recognizable;
• a second step for storing those of said patterns for which said classification results are evaluated as having low reliability and recognition of which is suspended by said first step in first storage means;
• a third step for storing those of said patterns for which said classification results are evaluated as having high reliability and said classification classes obtained by said first step in second storage means;
• a fourth step for optimizing said first step by using said patterns and said classification classes stored in said second storage means as a training sample;
• and
• a fifth step for determining said classification classes to which said patterns belong, by using said patterns stored in said first storage means after said first step is optimized by said fourth step.

A pattern recognition method comprising:

• a first step for storing a plurality of patterns in storage means and initializing the number of layers of a neural network, the number of neurons of each layer and a weight value of each neuron;
• a second step for obtaining a pattern recognition result of said patterns by said neural network;
• a third step for evaluating reliability of said pattern recognition result by the output value of said neural network;
• a fourth step for training said neural network;
• and
• a fifth step for storing said pattern recognition result in pattern recognition result storage means;
• wherein:
• (a) a step for inputting said patterns stored in said storage means to said neural network and the process of said second step;
• (b) the process of said fifth step;
• (c) the process of said third step;
• and
• (d) the process of said fourth step for executing training of said neural network by using said pattern recognition result of said patterns for which said classification results are evaluated as having high reliability as a desired output;
• are executed for all of said patterns stored in said pattern storage means, and after the processes from said step (a) to said step (d) are repeated until said pattern recognition result of all of said patterns is evaluated as having high reliability by said neural network or until a predetermined number of times is reached, said pattern recognition result stored in said pattern recognition result storage means is outputted.

A pattern recognition method comprising:

• a first step for classifying a plurality of inputted patterns to classification classes to which said patterns belong, evaluating reliability of the classification result, obtaining said classification classes for those of said patterns which provide said classification result having high reliability as recognizable, suspending recognition for those of said patterns for which said classification results have low reliability, and obtaining a first applicant and a second applicant of said classification classes to which said patterns are to be classified;
• a second step for counting the number of objects of said classification classes obtained by said first step for each classification class, and storing said first applicant and said second applicant obtained by said first step in storage means;
• and
• a third step for deciding said classification class of said pattern by using the number counted by said second step and said first applicant and said second applicant stored in said storage means.

An automated urinary sediment examination system comprising:

• a flow cell for forming a flat flow of a urine sample:
• laser beams to be irradiated to said flow cell;
• a particle detector for detecting the passage of solid components flowing through said flow cell;
• a pulse lamp for irradiating light to said flow cell in synchronism with the detection signal representing the passage of said solid components by said particle detector;
• a color television camera for imaging the images of said solid components as a still image;
• an image processor for classifying said solid components in said still image and measuring the amount of each kind of sediment components in said urine sample;
• and
• an output device for outputting the amount of each kind of said sediment components;
• said image processor including an A/D converter for converting analog image signals from said color television camera to digital image data, a region segmentation unit for segmenting said digital image data to a background region and an object region including said sediment components, a feature parameter calculation unit for calculating a feature parameter of said object region, a pattern recognition unit for executing recognition of said sediment components by inputting a pattern of said feature parameter as an input and classifying said sediment components to a classification item and a counter for counting the number of said sediment components classified to each of said classification items;
• wherein:
• said pattern recognition unit comprises:
• a first pattern recognition unit for classifying a plurality of said patterns inputted to classification classes to which said patterns belong, evaluating reliability of the classification result and outputting those of said patterns which provide said classification result having high reliability as recognizable to said classification class to which said pattern belong;
• a first storage unit for storing those of said patterns for which said classification results are evaluated as having low reliability and recognition of which is suspended by said first pattern recognition unit;
• a second storage unit for storing those of said patterns for which said classification results are evaluated as having high reliability by said first pattern recognition unit and said classification classes outputted by said first pattern recognition unit;
• a second recognition unit for inputting said patterns stored in said first storage unit and outputting said classification class to which said patterns belong by using said patterns and said classification class stored in said second storage unit as a training sample;
• and
• a pattern recognition method constitution unit for constructing said second pattern recognition unit by using said patterns and said classification classes stored in said second storage unit as a training sample.

An automated urinary sediment examination system comprising:

• a flow cell for forming a flat flow of a urine sample:
• laser beams to be irradiated to said flow cell;
• a particle detector for detecting the passage of solid components flowing through said flow cell;
• a pulse lamp for irradiating light to said flow cell in synchronism with a detection signal of the passage of said solid components by said particle detector;
• a color television camera for imaging the image of said solid components as a still image;
• an image processor for classifying said solid components in said still image and measuring the amount of each kind of sediment components in said urine sample;
• and
• an output device for outputting the amount of each kind of said sediment components;
• said image processor including an A/D converter for converting analog image signals from said color television camera to digital image data, a region segmentation unit for segmenting said digital image data into a background region and an object region including said sediment components, a feature parameter calculation unit for calculating a feature parameter of said object region, a pattern recognition unit for inputting patterns of said feature parameter, executing recognition of said sediment components and classifying them into classification classes, and a counter for counting the number of said sediment components classified into each of said classification items;
• wherein:
• said pattern recognition unit includes:
• a first pattern recognition unit being set to an initial state before a plurality of said patterns are inputted, for classifying said patterns to classification classes to which said patterns belong, evaluating reliability of the classification result, and outputting the classification class to which said patterns belong, for those of said patterns which provide the classification result having high reliability as recognizable;
• a first storage unit for storing those of said patterns for which said classification results are evaluated as having low reliability and recognition of which is suspended by said first pattern recognition unit;
• a second storage unit for storing those of said patterns for which said classification results are evaluated as having high reliability by said first pattern recognition unit and said classification classes outputted by said first pattern recognition unit;
• a pattern recognition method adjustment unit for optimizing said first pattern recognition unit by using said patterns and said classification classes stored in said second storage unit as a training sample;
• and
• a reference pattern recognition method storage unit for storing an initial state of said first pattern recognition unit;
• and
• wherein:
• said patterns stored in said first storage unit is inputted to said first pattern recognition unit after said first pattern recognition unit is optimized by said pattern recognition method adjustment unit, and said classification classes to which said patterns belong are outputted.

An automated urinary sediment examination system comprising:

• a flow cell for forming a flat flow of a urine sample:
• laser beams to be irradiated to said flow cell;
• a particle detector for detecting the passage of solid components flowing through said flow cell;
• a pulse lamp for irradiating light to said flow cell in synchronism with a detection signal of the passage of said solid components by said particle detector;
• a color television camera for imaging the images of said solid components as a still image;
• an image processor for classifying said solid components in said still image and measuring the amount of each kind of said sediment components in said urine sample;
• and
• an output device for outputting the amount of each kind of said sediment components;
• said image processor including an A/D converter for converting analog image signals from said color television camera to digital image data, a region segmentation unit for segmenting said digital image data into a background region and an object region including said sediment components, a feature parameter calculation unit for calculating a feature parameter of said object region, a pattern recognition unit for inputting the patterns of said feature parameter, executing recognition of said sediment components and classifying them to classification items and a counter for counting the number of said sediment components classified into each of said classification items;
• wherein:
• said pattern recognition unit includes:
• a pattern storage unit for storing a plurality of patterns;
• a neural network for executing pattern recognition;
• a reliability evaluation unit for evaluating reliability of a pattern recognition result of said patterns from the output value of said neural network;
• an initial weight value storage unit for storing the number of layers of said neural network in an initial state, the number of neurons of each of said layers and a weight value of each of said neurons;
• neural network training means for executing training of said neural network;
• and
• a pattern recognition result storage unit for storing said pattern recognition result;
• and
• wherein:
• after a plurality of said patterns are stored in said pattern storage unit and said neural network is initialized in accordance with the content of said initial weight value storage unit;
• (a) a process for executing pattern recognition by inputting said patterns stored in said pattern storage unit to said neural network;
• (b) a process for storing said pattern recognition result in said pattern recognition result storage unit;
• (c) a process for evaluating reliability of said pattern recognition result by said reliability evaluation unit by using the output value of said neural network;
• and
• (d) a process for executing training of said neural network by said neural network training means by using said pattern recognition result of said patterns for those of said patterns for which said classification results are evaluated as having high reliability;
• are executed for all of said patterns stored in said pattern storage unit, and after said process (a) to said process (d) are repeated until reliability of said pattern recognition result of all of said patterns is evaluated as high by said neural network or until a predetermined number of times is reached, said pattern recognition result stored in said pattern recognition result storage unit is outputted.

An automated urinary sediment examination system comprising:

• a flow cell for forming a flat flow of a urine sample;
• laser beams to be irradiated to said flow cell;
• a particle detector for detecting the passage of solid components flowing through said flow cell;
• a pulse lamp for irradiating light to said flow cell in synchronism with a detection signal of the passage of said solid components by said particle detector;
• a color television camera for imaging the images of said solid components as a still image;
• an image processor for classifying said solid components in said still image and measuring the amount of each kind of sediment components in said urine sample;
• and
• an output device for outputting the amount of each kind of said sediment components;
• said image processor including an A/D converter for converting analog image signals from said color television camera to digital image data, a region segmentation unit for segmenting said digital image data to a background region and an object region including said sediment components, a feature parameter calculation unit for calculating a feature parameter of said object region, a pattern recognition unit for executing recognition of said sediment components by inputting patterns of said feature parameter and classifying them into classification items and a counter for counting the number of said sediment components classified into each of said classification items;
• wherein:
• said pattern recognition unit includes:
• a first pattern recognition unit for classifying a plurality of said patterns inputted to classification classes to which said patterns belong, evaluating reliability of the classification result, outputting said classification class as recognizable for those of said patterns which provide said classification result having high reliability, suspending recognition for those of said patterns for which said classification results have low reliability, and outputting a first applicant and a second applicant of said classification class to which said patterns are to be classified;
• a storage unit for storing said first applicant and said second applicant;
• a counter for counting the number of objects for which said classification class is outputted by said first pattern recognition unit, for each classification class;
• and
• reasoning means for deciding said classification class of said patterns by using the content of said counter and said first applicant and said second applicant stored in said storage unit.

An automated urinary sediment examination method including:

• a step of forming a flat flow of a urine sample in a flow cell;
• a step of irradiating laser beams to said flow cell;
• a step of detecting the passage of solid components flowing through said flow cell;
• a step of irradiating pulse lamp light to said flow cell in synchronism with the detection of the passage of said solid components;
• a step of imaging the image of said solid components as a still image by a color television camera;
• a step of classifying said solid components in said still image and measuring the amount of each kind of sediment components in said urine sample;
• and
• a step of outputting the amount of each kind of said sediment components;
• said image processing step including:
• a step of converting analog image signals from said color television to digital image data;
• a step of segmenting said digital image data to a background region and an object region including said sediment components;
• a step of calculating a feature parameter of said object region;
• a step of executing pattern recognition to recognize said sediment components and to classify them into classification items;
• and
• a step of counting the number of said sediment components classified to each of said classification items;
• wherein:
• said automated urinary sediment examination method comprises:
• a first step of classifying a plurality of said patterns to classification classes to which said patterns belong, evaluating reliability of the classification result and obtaining the classification class to which said patterns belong, as recognizable for those of said patterns which provide said classification result having high reliability;
• a second step of storing those of said patterns for which said classification results are evaluated as having low reliability and recognition of which is suspended in a first storage unit;
• a third step of storing said patterns for which said classification results are evaluated as having high reliability by said first step and said classification class in a second storage unit;
• a fourth step of executing second pattern recognition to obtain said classification class to which said patterns belong, from said patterns stored in said first storage unit by using said patterns and said classification class stored in said second storage unit as a training sample;
• and
• a fifth step of constructing said second pattern recognition by using said patterns and said classification class stored in said second storage unit as a training sample.

An automated urinary sediment examination method including:

• a step of forming a flat flow of a urine sample in a flow cell;
• a step of irradiating laser beams to said flow cell;
• a step of detecting the passage of solid components flowing through said flow cell;
• a step of irradiating pulse lamp light to said flow cell in synchronism with the detection of the passage of said solid components;
• a step of imaging the image of said solid components as a still image by a color television camera;
• a step of executing an image processing for classifying said solid components in said still image and measuring the amount of each kind of sediment components in said urine sample;
• and
• a step of outputting the amount of each kind of said sediment components;
• said image processing step including:
• a step of converting analog image signals from said color television camera to digital image data;
• a step of segmenting said digital image data into a background region and an object region including said sediment components;
• a step of calculating a feature parameter of said object region;
• and
• a step of executing pattern recognition to recognize said sediment components and to classify them into classification items by inputting the patterns of said feature parameter;
• wherein:
• said step of executing pattern recognition includes:
• a first step being set to an initial state before a plurality of said patterns are inputted, for classifying said patterns to classification classes to which said patterns belong, evaluating reliability of the classification result, and obtaining the classification class to which said patterns belong, as recognizable for those of said patterns which provide the classification result having high reliability;
• a second step of storing those of said patterns for which said classification results are evaluated as having low reliability by said first step and recognition of which is suspended, in a first storage unit;
• a third step of storing those of said patterns for which said classification results are evaluated as having high reliability by said first step and said classification class obtained by said first step in a second storage unit;
• a fourth step of optimizing said first step by using said patterns and said classification classes stored in said second storage unit as a training sample;
• and
• a fifth step of deciding said classification class to which said patterns belong, by using said patterns stored in said first storage unit after said first step is optimized by said fourth step.

An automated urinary sediment examination method including:

• a step of forming a flat flow of a urine sample in a flow cell;
• a step of irradiating laser beams to said flow cell;
• a step of detecting the passage of solid components flowing through said flow cell;
• a step of irradiating pulse lamp light to said flow cell in synchronism with the detection of the passage of said solid components;
• a step of imaging the image of said solid components as a still image by a color television camera;
• a step of executing image processing for classifying said solid components in said still image and measuring the amount of each kind of sediment components in said urine sample;
• and
• a step of outputting the amount of each kind of said sediment components;
• said image processing step including:
• a step of converting analog image signals from said color television camera to digital image data;
• a step of segmenting said digital image data into a background region and an object region including said sediment components;
• a step of calculating a feature parameter of said object region;
• a step of executing pattern recognition to recognize said sediment components by using the patterns of said feature parameter as an input and classifying them into classification items;
• and
• a step of counting the number of said sediment components classified into each of said classification items;
• wherein:
• said pattern recognition step includes:
• a first step of storing a plurality of said patterns in a storage unit, and initializing the number of layers of a neural network, the number of neurons of each layer and a weight value of each neuron;
• a second step of obtaining a pattern recognition result of said patterns by said neural network;
• a third step of evaluating reliability of said pattern recognition result of said patterns by an output value of said neural network;
• a fourth step of executing training of said neural network;
• and
• a fifth step of storing said pattern recognition result in a pattern recognition result storage unit;
• and
• wherein:
• (a) a processes of a step of inputting said patterns stored in said storage unit to said neural network and said second step;
• (b) a process of said fifth step;
• (c) a process of said third step;
• and
• (d) a process of said fourth step of executing training of said neural network by using said pattern recognition result of said patterns for which said classification results are evaluated as having high reliability as a desired output;
• are executed for all of said patterns stored in said storage unit, and after the processes from said item (a) to said item (d) are repeated until reliability of said pattern recognition result of all of said patterns is evaluated as high by said neural network or until a predetermined number of times is reached, said pattern recognition result stored in said pattern recognition result storage unit is outputted.

An automated urinary sediment examination method including:

• a step of forming a flat flow of a urine sample in a flow cell;
• a step of irradiating laser beams to said flow cell;
• a step of detecting the passage of solid components flowing through said flow cell;
• a step of irradiating pulse lamp light to said flow cell in synchronism with the detection of the passage of said solid components;
• a step of imaging the image of said solid components as a still image by a color television camera;
• a step of executing an image processing for classifying said solid components in said still image and measuring the amount of each kind of sediment components in said urine sample;
• and
• a step of outputting the amount of each kind of said sediment components;
• said image processing step including:
• a step of converting analog image signals from said color television camera to digital image data;
• a step of segmenting said digital image data into a background region and an object region including said sediment components;
• a step of calculating a feature parameter of said object region;
• a step of executing pattern recognition to recognize said sediment components by using the patterns of said feature parameter as an input and to classify them into classification items;
• and
• a step of counting the number of said sediment components classified to each of said classification items;
• wherein:
• said pattern recognition step includes:
• a first step of classifying a plurality of said patterns inputted to classification classes to which said patterns belong, evaluating reliability of the classification result, obtaining said classification class as recognizable for those of said patterns which provide said classification result having high reliability, suspending recognition for those of said patterns for which said classification results have low reliability, and obtaining a first applicant and a second applicant of said classification class to which said patterns are to be classified;
• a second step of counting the number of objects of said classification classes obtained by said first step for each of said classification classes, and storing said first applicant and said second applicant obtained by said first step in a storage unit;
• and
• a third step of deciding said classification class of said patterns by using the number counted by said second step and said first applicant and said second applicant stored in said storage unit.