how to fix the following error-

2 次查看（过去 30 天）

Annie micheal 2017-10-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/362082-how-to-fix-the-following-error

评论： Walter Roberson 2017-10-26

I'm using matlab 2015b version. i ve constructed a Multilayer perceptron model to predict the gender of a person. While testing the model i get the following error-Systems of struct class cannot be used with the "predict" command. Convert the system to an identified model first, such as by using the "idss" command. Error in maintest (line 55) label = predict(model,S); how to fix this error..

14 个评论
显示 12更早的评论隐藏 12更早的评论

Annie micheal 2017-10-20

编辑：Torsten 2017-10-20

在 MATLAB Online 中打开

clear all;
clc;
modelFile = 'model_frontal.mat';
[fname,pname] = uigetfile('*.*','Select the input image file');
filename = sprintf('%s%s',pname,fname); 
img1=imread(filename);
load(modelFile, 'npdModel');
rects = DetectFace(npdModel, img1);
numFaces = length(rects);    
%fprintf('%d faces detected.\n', numFaces);
if numFaces > 0
    border = round(size(img1,2) / 300);
    if border < 2, border = 2; end
    for j = 1 : numFaces
        I = DrawRectangle(img1, rects(j).col, rects(j).row, rects(j).width, rects(j).height, [0 255 0], border);
        cropI = imcrop(I,[rects(j).col, rects(j).row, rects(j).width, rects(j).height]);
    end
end
sexs=zeros(size(rects, 1),1);
n=size(rects, 1);
text_str = cell(n,1);
position = zeros(n,2);
for i=1:n
   text_str{i} = ['male '];
end
for i = 1:n
    xs = rects(j).col;
    xe = xs + rects(j).width;
       ys = rects(j).row;
   ye = ys + rects(j).height;
    gray=rgb2gray(I(ys:ye, xs:xe, :));
    gray2=imresize(gray,[256 256]);
  %  load radius1
   class = 0;
   value=WLD(gray2)
O = value;
%img= imresize(img,[256 256]);
%roi = [rects(j).col; rects(j).row; rects(j).width; rects(j).height];
roi= [1; rects(j).height;1;rects(j).width;];
%Q =mean(Q2);
 S=[O]
 %#function model
modelFile = 'MainClassifier.mat';
   load(modelFile, 'model');
  sys=idss(model);   
end
%sys=idss(model);
   label = predict(sys,S);
    position(i,1)=xs;
    position(i,2)=ye;
        if label==1
            text_str='male';
         else
             text_str='female';
        end
shapeInserter = vision.ShapeInserter('BorderColor','Custom','CustomBorderColor',[255 50 50]);
I = step(shapeInserter, img1,int8([rects(j).col, rects(j).row, rects(j).width, rects(j).height]));
A = insertText(I, position, text_str,'FontSize',18,'BoxOpacity',0.4,'TextColor','white');
%RGB = insertText(I,position,text_str,'FontSize',18,'BoxColor',box_color,'BoxOpacity',0.4,'TextColor','white');
figure;
imshow(A);

Annie micheal 2017-10-25

编辑：Walter Roberson 2017-10-25

在 MATLAB Online 中打开

from thrid- party contribution.. i ve posted the code

function [model cc output] = train_mlp(input, target, hidden, iterations, learning_rate, momentum)
    % this is the function that handles all the looping and running of the
    % neural network, it initializes the network based on the number of
    % hidden layers, and presents every item in the input over and over,
    % $iterations$ times.
    % hidden is the only complicated variables.  Like weka, it accepts a
    % list of values (as a row vector), and interprets it as the number of
    % neurons in each hidden layer, so [2 2 2] means there will be an input
    % layer defined by the size of the input, three hidden layers, with 2 
    % neurons each, and the output layer, defined by the target.  I think
    % it's pretty good.
      % initialize the output
      model = [];
      model.input=input;
      model.target=target;
      model.learning_rate = learning_rate;
      model.momentum = momentum; % for some heavy ball action
      % characterize the input and output
      [ntrain nInLayer] = size(input);
      [jnk nOutLayer] = size(target);
      % keep track of how many neurons are in each layer
      nNeurons = [nInLayer hidden nOutLayer];
      nNeurons(nNeurons == 0) = []; % remove 0 layers, to allow putting a zero for no hidden layers
      % there are one fewer sets of weights and biases as total layers
      nTransitions = length(nNeurons)-1;
      for i = 1:nTransitions % initialize the weights between layers, and the biases (past the first layer)
          model.weights{i} = randn(nNeurons(i),nNeurons(i+1)); 
          % the weight matrix has X rows, where X is the number of input
          % neurons to the layer, and Y columns, where Y is the number of
          % output neurons.  multiplication of the input with the weight
          % matrix transforms the dimensionality of the input to that of the
          % output.  Initialization is done here randomly.
          model.biases{i} = randn(1,nNeurons(i+1));
          %double(model.biases{i});
          % biases are random as well
    model.lastdelta{i} = 0; 
      end
      for i = 1:iterations % repeat the whole training set over and over
          order = randperm(ntrain);  % randomize the presentations
          for j = 1:ntrain
              % update_mlp is where the training is actually done
              model = update_mlp(model, input(order(j),:), target(order(j),:));
          end
      end
      % test the performance on the training set after all is trained.
      [output cc] = test_mlp(model, input, target);
      save MainClassifier.mat model;
  end

Annie micheal 2017-10-26

编辑：Walter Roberson 2017-10-26

my predict() command invoke https://in.mathworks.com/help/nnet/ref/predict.html

Walter Roberson 2017-10-26

Okay, that is progress.

The Neural Network predict() command requires either a

SeriesNetwork or a DAGNetwork (Directed acyclic graph)

When I examine articles about MLP, I find that they are a non-cascaded Feed Forward Network, such as one might create with feedforwardnet() . However, feedforwardnet() is used for function approximation and clustering, and predict() is not applicable to such networks. Considering your purpose, I wonder if you would be better off with a <https://www.mathworks.com/help/nnet/ref/patternnet.html patternnet() for pattern recognition? Those do not use predict() either though.

The predict() for SeriesNetwork deals with networks that look fairly different in structure that MLP would have -- although possibly you could use such a network, it looks much more complicated than you need to me.

Anyhow, you have the fundamental problem that the third-party MLP training routine that you are using does not create a network structure that is in the form expected by any of the Neural Network Toolbox functions. If we could narrow down your needs, then possibly we could extract information from the training you did and use it to build something that could be used with the Neural Network toolbox.

... but I suspect that what you are supposed to do with the training structure you created is to call a third party routine to use it with. Certainly that would be the case if you are using http://3options.net/brainannex/multilayer-perceptron-in-matlab-octave/

请先登录，再进行评论。

请先登录，再回答此问题。