Fig. 6

The structure of the proposed object detection network. a The detection network consists of three main parts: The (1) Backbone, a replaceable convolutional neural network for clustering and forming image features from fine and coarse gained images; The (2) Neck, a series of network layers for fusing and combining image features which are then sent to the predicting network, and (3), The Output, a network for prediction of image features, generation of bounding boxes and prediction of results. b The sub-module components of the detection network: (1) Convolutional layer. (2) Batch normalization operation. (3) Leaky Relu activation function. (4) Slicing operation. (5) Concatenate function puts slices into a block with 4X channels