Implementation of the MKMS-FPN backbone. Feature maps from backbone stages C2-C4 are projected and fused through top-down and lateral connections to form a multi-scale pyramid. The SPFF (spatial pyramid feature fusion) module aggregates multi-scale context using multiple pooling branches followed by convolutional fusion. The FCM (feature channel module) applies spatial and channel attention on two branches that are recombined to produce attention-weighted features. The MKPConv block performs parallel depthwise convolutions with kernel sizes {1,3,5,7} (with padding) followed by a 1 × 1 pointwise convolution to fuse multi-scale responses. The resulting pyramid features (P2-P5) are used for region proposal and detection.