Traditional Land-use classification techniques relied on spectral information. I would suggest to do post classification refinement using knowledge based or decision tree approach to eliminate the misclassification errors. Have a look at this paper, it might be helpful to you,

http://www.sciencedirect.com/science/article/pii/S1110982315000381

Good luck

Recommended paper: "Death to Kappa: birth of quantity disagreement and allocation disagreement for accuracy assessment", by Robert Gilmore Pontius Jr & Marco Millones: 

http://www.tandfonline.com/doi/abs/10.1080/01431161.2011.552923

Cheers

AG

Please check the page below. It may allow you visually comparison as well as statistically.

http://mck.riks.nl/

Also please check this article below. I personally found it very interesting.

Integrating User Needs on Misclassification Error Sensitivity into Image Segmentation Quality Assessment

http://www.sciencedirect.com/science/article/pii/S0099111215301373

R toolbox, just see below

http://www.ciga.unam.mx/sw/accurassesEN.html

There are various points which can improve the accuracy of landuse mapping using satellite images. First, the most important thing is to have a good and high quality data. both spatial and spectral resolution are important for accurate mapping. in terms of methods, based on recent literature, SVM and object based methods including optimization of segmentation and classification as well as the transferability of the rules are the future direction. However, the recent literature lacking the development of transferable rules because they mainly focus on data and they need to find a theoretical approach to improve the transferability. For example, using only significant bands and significant attributes are very important to achieve transferable rules. To conclude, both data and feature extraction methods need to be improved to achieve high quality maps.