Improvement of Multiclass Classification of Pavement Objects Using Intensity and Range Images

Table 3

Comparison of deep CNNs for classification of pavement objects using multiscale intensity and range input tiles.

Metric	Method	Input image	Crack	Crack seal	Patch	Pothole	Marker	Manhole	Curbing	Asphalt	Avg
Precision	M-VGG16	Intensity	0.755	0.850	0.874	0.865	0.949	0.930	0.982	0.977	0.898
		Intensity + Range	0.824	0.966	0.929	0.920	0.958	0.933	0.990	0.982	0.938
	M-VGG19	Intensity	0.775	0.883	0.859	0.902	0.959	0.959	0.984	0.977	0.912
		Intensity + Range	0.786	0.904	0.92	0.914	0.959	0.910	0.986	0.985	0.921
	M-ResNet50	Intensity	0.772	0.894	0.798	0.885	0.952	0.952	0.976	0.975	0.901
		Intensity + Range	0.781	0.940	0.932	0.898	0.949	0.917	0.980	0.984	0.923
	M-DenseNet121	Intensity	0.756	0.839	0.869	0.883	0.960	0.934	0.981	0.975	0.900
		Intensity + Range	0.811	0.889	0.972	0.920	0.961	0.933	0.983	0.984	0.932
	DACNN (ours)	Intensity	0.864	0.897	0.965	0.947	0.966	0.919	0.983	0.986	0.941
		Intensity + Range	0.887	0.942	0.972	0.953	0.971	0.965	0.993	0.987	0.959
Recall	M-VGG16	Intensity	0.640	0.771	0.838	0.854	0.965	0.910	0.985	0.984	0.868
		Intensity + Range	0.699	0.831	0.954	0.919	0.967	0.946	0.988	0.989	0.912
	M-VGG19	Intensity	0.632	0.773	0.883	0.873	0.959	0.906	0.985	0.986	0.875
		Intensity + Range	0.758	0.920	0.952	0.920	0.961	0.966	0.989	0.985	0.931
	M-ResNet50	Intensity	0.606	0.617	0.853	0.775	0.961	0.897	0.985	0.984	0.835
		Intensity + Range	0.735	0.693	0.918	0.891	0.970	0.927	0.989	0.985	0.889
	M-DenseNet121	Intensity	0.636	0.766	0.824	0.810	0.952	0.906	0.984	0.985	0.858
		Intensity + Range	0.741	0.929	0.917	0.943	0.962	0.946	0.990	0.988	0.927
	DACNN (ours)	Intensity	0.780	0.837	0.942	0.909	0.957	0.937	0.990	0.991	0.918
		Intensity + Range	0.805	0.947	0.958	0.956	0.966	0.937	0.990	0.993	0.944
F-score	M-VGG16	Intensity	0.693	0.808	0.856	0.859	0.958	0.920	0.984	0.981	0.882
		Intensity + Range	0.756	0.893	0.941	0.920	0.961	0.940	0.989	0.985	0.923
	M-VGG19	Intensity	0.696	0.824	0.871	0.887	0.959	0.931	0.985	0.982	0.892
		Intensity + Range	0.772	0.912	0.936	0.917	0.960	0.937	0.988	0.985	0.926
	M-ResNet50	Intensity	0.679	0.730	0.825	0.827	0.956	0.924	0.980	0.980	0.863
		Intensity + Range	0.757	0.797	0.925	0.895	0.960	0.922	0.984	0.985	0.903
	M-DenseNet121	Intensity	0.691	0.801	0.846	0.845	0.956	0.919	0.982	0.980	0.878
		Intensity + Range	0.774	0.908	0.944	0.932	0.962	0.940	0.986	0.986	0.929
	DACNN	Intensity	0.820	0.866	0.953	0.928	0.961	0.928	0.986	0.988	0.929
		Intensity + Range	0.844	0.944	0.965	0.954	0.969	0.951	0.992	0.990	0.951