In recent years, object detection has been widely developed, and small object detection in object detection has received more and more attention. Feature pyramid networks are often used in object detection. Because the feature pyramid network can be detected on feature maps of different scales, more objects can be detected using the feature pyramid network for detection. However, in the pyramid feature network, due to the reduction of channels in feature fusion, the high-level features of the pyramid will lose the detailed information of the object. At the same time, because the background information at the bottom of the pyramid is complicated, it is not conducive to detecting small object ions. In order to better improve the performance of feature pyramid detection for small objects, we propose an object detection method based on residual branch (SORB), which improves the detection accuracy of small objects while maintaining the existing object detection accuracy. Our method improves the network structure of the traditional feature pyramid. We also recalculated the weights of the network to reduce the semantic gap between different features in the feature pyramid. Our method is validated on the VOC2012 dataset, and the experimental results show that our method has good results

Clinoptilolite were collected from the quarries of Ajanta caves , Marathwada    (Maharashtra ). Sample was crushed and sieved to get 106 μm sized crystals for the ion exchanged. The sample is treated with 1 M solution of Ammonium Nitrate with stirring at 95°C. for the six hours to get NH4 ion exchanged form of Clinoptilolite ,again heated at 250°C for 48 hours for getting H – Clinoptilolite. Characterization was made using XRD,IR at NCL Pune  Dielectric study was made using LCR Bridge. Pellets of  H- Clinoptilolite were prepared..Variation of dielectric constant, dielectric loss, dielectric conductivity and relaxsession time were measured from 20Hz to 20KHz