Otsu's Threshold Question

I am attempting to create a classifier which detects the solar panels in a series of images. To create the training data i have been using Otsu's threshold to seperate the blue solar cells from the white grids (see first image below). I have been able to isolate the solar panels from the background, as seen in the second image, and then use the threshold to seperate the classes. I, however, want the blue cells to be highlighted instead of the white grid lines. Does anyone know how to do this? Below is also the code I have been running and the output images at various points.

RGB = imread('1-108.png'); %input image (as seen above)
GT = imread('1-108gt.png'); %original groundtruth of input image which highlights both grid and cell
I = rgb2gray(RGB); %convert RGB to intensity

[row,col] = size(I);
com = uint8(zeros(size(I)));
%when the groundtruth is logic 1, replace this with the corresponding intensity pixel
for r=1:row
    for c=1:col
        if GT(r,c)==1 
            com(r,c) = I(r,c);
        else
            com(r,c) = 0;
        end
    end
end

figure
imshow(com)

[counts, x] = imhist(com,16);
figure
stem(x,counts)

T = otsuthresh(counts(2:16,:)); %does not consider the first column as this is all the background pixels
BW = imbinarize(com,T);
figure
imshow(BW)

Answer 1

[Matt J]

I= double(rgb2gray(imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/729084/1-108.png')));
GT = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/729089/1-108gt.png');

com= uint8((GT==1).*I);

[counts, x] = imhist(com,16);
T = otsuthresh(counts(2:16,:)); %does not consider the first column as this is all the background pixels
BW =GT-imbinarize(com,T);

BW=imclose(BW,ones(3,1)); 
BW=imfill(BW,'holes');
BW=imopen(BW,ones(5,1));
BW=imopen(BW,ones(1,3));


imshow(BW)

[Petrus van Aswegen]

Genius mate, thanks heaps