Olution operation to acquire the Query, Key, and Worth branches. Right after getting into the Q branch, the Guretolimod Agonist Function map with aa size of C HW was flatbranches. Immediately after getting into the Q branch, the function map with size of C H W was flattened intotwo-dimensional vector with a size of of N, where N =N = eature map tened into a a two-dimensional vector having a size C C N, exactly where H W. W. Function map Q was transposed to obtain a feature vector Q’ with a size of N C. Immediately after the function Q was transposed to get a function vector Q’ with a size of N C. Following the feature map map entered branch K, the function map having a size of C H W was obtained by means of entered branch K, the function map having a size of C H W was obtained via spatial spatial FM4-64 supplier pyramid pooling to achieve a reduction in dimensionality. The spatial pyramidRemote Sens. 2021, 13, 4532 Remote Sens. 2021, 13, x FOR PEER REVIEW6 of 20 6 ofpooling operationto reach a reduction in dimensionality. The spatial pyramid module pyramid pooling is shown in Figure 5 below. The spatial pyramid pooling pooling performedis shown in Figure 5 beneath. The spatial pyramid having a window size of n nthe operation the maximum pooling in the input function map pooling module performed to get the feature map the input function map n. Theafeature map with n size of C n n maximum pooling of using a size of C n with window size of a n to acquire the was made use of to represent the sampling final results of representative anchor points in each and every location of function map having a size of C n n. The function map having a size of C n n was used towards the origin feature map. Then, all the function maps following the spatial pyramid pooling had been represent the sampling outcomes of representative anchor points in every location on the origin flattened and concatenated to get a feature vector using a size of C S, exactly where S was feature map. Then, all of the feature maps following the spatial pyramid pooling were flattened determined by the size and quantity of the selected pooling windows. By way of example, in this and concatenated to obtain a feature vector having a size of C S, where S was determined article, the pooling widow is 1 1, 3 three, 6 6, and 8 8, and S is equal to: by the size and variety of the chosen pooling windows. One example is, within this short article, the pooling widow is 1 1, three three, six 6, = eight eight, and =is equal to: S and n2 S=n1,3,6,8 , , , =Figure five. Structure of spatial pyramid pooling. Figure five. Structure of spatial pyramid pooling.Following the function map, X entered the Query and Important branches, and the function vectors Immediately after the feature map, X entered the Query and Crucial branches, plus the feature vectors Q’ with a size of N C and K’ using a size of C S are matrix multiplied to receive function Q’ having a size of N C and K’ using a size of C S are matrix multiplied to acquire feature map QK’. Function map QK’ was normalized by SoftMax to get the consideration map QK. map QK’. Function map QK’ was normalized by SoftMax to get the attention map QK. The goal of this was to calculate the connection involving each pixel in feature vector The objective of this was to calculate the connection among every pixel in feature vector Q’ and every pixel in K’. Within this way, we can acquire a feature map of C S size, which Q’ and every pixel in K’. Within this way, we are able to acquire a feature map of C S size, which represents the consideration partnership among the Query pixel along with the feature anchor point represents the attention relationship in between the Query pixel and the feature anchor point inside the Essential, and repres.
