How do I create an open class classifier
3 次查看(过去 30 天)
显示 更早的评论
I am calling function kNN_model_add_class to trained and create one class call dogs.
How do I create an 'Open class' where there will be all the other sounds(i do not want to specify them just all the other)
I am trying to make One vs all classifier.. Dog or no Dog
Thanks
strDir = 'C:\Users\User\Desktop\3rd year\Nahravky\cats_dogs';
Statistics = {'mean', 'median', 'std', 'stdbymean', 'max', 'min'};
stWin = 0.040;
stStep = 0.040;
mtWin = 2;
mtStep = 1;
kNN_model_add_class('Mfcctrain.mat', 'dogs', [ strDir '\trainpes'], ...
Statistics, stWin, stStep, mtWin, mtStep);
function kNN_model_add_class
function kNN_model_add_class(modelName, className, classPath, ...
listOfStatistics, stWin, stStep, mtWin, mtStep)
%
% function kNN_model_add_class(modelName, className, classPath, ...
% listOfStatistics, stWin, stStep, mtWin, mtStep)
%
% This function adds an audio class to the kNN classification model
%
% ARGUMENTS;
% - modelName: the filename of the model (mat file)
% - className: the name of the audio class to be added to the model
% - classPath: the path of the directory where the audio segments of the
% new class are stored
% - listOfStatistics: list of mid-term statistics (cell array)
% - stWin, stStep: short-term window size and step
% - mtWin, mtStep: mid-term window size and step
%
% Example:
% kNN_model_add_class('modelSpeech.mat', 'speech', './Music/', ...
% {'mean','std',}, 0.050, 0.025, 2.0, 1.0);
%
if ~exist(classPath,'dir')
error('Audio sample path is not valid!');
else
classPath = [classPath filesep];
end
% check if the model elaready exists:
fp = fopen(modelName, 'r');
if fp>0 % check if file already exists
load(modelName);
end
% Feature extraction:
D = dir([classPath '*.wav']);
F = [];
for (i=1:length(D)) % for each wav file in the given path:
curFileName = [classPath D(i).name];
FileNamesTemp{i} = curFileName;
% mid-term feature extraction for each wav file:
[data, fs] = audioread(curFileName);
signal = struct('Filt_data', data, 'SampleRate', fs);
midFeatures = featureExtractionFile(signal, ...
stWin, stStep, mtWin, mtStep, listOfStatistics);
% midFeatures = featureExtractionFile(curFileName, ...
% stWin, stStep, mtWin, mtStep, listOfStatistics);
% long-term averaging:
longFeatures = mean(midFeatures,2);
F = [F longFeatures];
end
% save the model:
Statistics = listOfStatistics;
fp = fopen(modelName, 'r');
if fp<0 % model does not exist --> generate
ClassNames{1} = className;
Features{1} = F;
FileNames{1} = FileNamesTemp;
save(modelName, 'ClassNames', 'Features', ...
'Statistics', 'stWin', 'stStep', 'mtWin', 'mtStep', 'FileNames');
else
load(modelName);
ClassNames{end+1} = className;
Features{end+1} = F;
FileNames{end+1} = FileNamesTemp;
save(modelName, 'ClassNames', 'Features', ...
'Statistics', 'stWin', 'stStep', 'mtWin', 'mtStep', 'FileNames');
end
6 个评论
Walter Roberson
2020-11-18
You do not create an open class. A specific open class would have to be trained as having particular characteristics, and anything that did not have those characteristics would not match.
Instead you define a threshold for matching each of your defined classes, and if the image does not meet the threshold for any of the defined classes then you say that it is Other.
回答(0 个)
另请参阅
类别
在 Help Center 和 File Exchange 中查找有关 Classification 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!