Extracting features from pretrained network and feed them into a softmax layer for calssification

I use ResNet50 to extract features from images then I want to feed these features into softmax layer for classification but I don't know how to feed these features into softmax layer. I used fitcecoc for classification but it didn't give me the good results for that reason I want to use softmax layer rather than fitcecoc.

Here is my code:

imds = imageDatastore('chromosomes','IncludeSubfolders',true,'LabelSource','foldernames');

[trainingSet,testSet] = splitEachLabel(imds,0.8, 'randomize');

net = resnet50 ();

imageSize = net.Layers(1).InputSize;

augmentedTrainingSet = augmentedImageDatastore(imageSize, trainingSet,'ColorPreprocessing','gray2rgb');

augmentedTestSet = augmentedImageDatastore(imageSize, testSet,'ColorPreprocessing','gray2rgb');

featureLayer = 'activation_43_relu';

trainingFeatures = activations(net, augmentedTrainingSet, featureLayer, 'OutputAs','rows');

testFeatures = activations(net, augmentedTestSet, featureLayer,'OutputAs','rows');

trainingLabels = trainingSet.Labels;

testLabels = testSet.Labels;

classifier = fitcecoc(trainingFeatures,trainingLabels);

YPred = predict(classifier,testFeatures);

accuracy = mean(YPred == testLabels)