Fixed ANN_MLP dw initialization when moment is not 0

modules/ml/src/ann_mlp.cpp

@@ -739,7 +739,7 @@ public:
             int n = layer_sizes[i];
             x[i].resize(n+1);
             df[i].resize(n);
-            dw[i].create(weights[i].size(), CV_64F);
+            dw[i] = Mat::zeros(weights[i].size(), CV_64F);
         }
 
         Mat _idx_m(1, count, CV_32S);