Currently, the field of remote sensing object detection has made significant strides, yet effectively detecting both conventional rectangular objects and distinctive slender objects simultaneously remains a persistent challenge, owing to their inherent coexistence in aerial imagery with markedly different geometric characteristics. Here, rectangular objects refer to those with nearly rectangular bounding boxes, while slender objects denote those with elongated rectangular bounding boxes in detection tasks. This paper demonstrates that rectangular and slender objects exhibit mutual dependency in detection models, necessitating joint consideration of both object types. Therefore, inspired by large-kernel square convolution and strip convolution operations, we investigate their synergistic integration and propose a novel detection network architecture that effectively combines both convolution types. Square convolution operations excel at capturing features of approximately square-shaped objects, while strip convolution operations are particularly effective for extracting characteristics of slender objects, thereby collectively enhancing the overall detection capability. Our method demonstrates superior performance on the DOTA-v1.5 dataset, achieving a single-scale mAP of 72.10% and outperforming several classical detection models.