improve the performance of semantic segmentation

i used this options for training SegNet on my own data (medical image) but i got high error and looowww accuracy

i reduced the value of learning rate and reguralization but still the same problem..what is your advice to get loww error high accuracy

options = trainingOptions('sgdm', 'Momentum', 0.9,'InitialLearnRate', 0.001, 'LearnRateSchedule', 'piecewise', ...

'LearnRateDropFactor', 0.9,'LearnRateDropPeriod',1,'L2Regularization', 0.0001,'MaxEpochs', 30, ...

'MiniBatchSize', 4, 'Shuffle', 'every-epoch', 'Plots','training-progress', 'VerboseFrequency', 10);