In recent years, the rapid development of deep learning has significantly improved the performance of joint source-channel coding (JSCC) in wireless image transmission. However, the current channel adaptive mechanism adopts one-size-fits-all strategy for features in different layers, making it difficult to satisfy the differentiation between the low-level details and the high-level semantics under the varying channel conditions. In addition, most existing methods extract only features in spatial domain, ignoring frequency domain which is important to wireless image transmission. In this paper, we propose a novel end-to-end Frequency Spatial Hybrid JSCC model incorporating a Channel Adaptive Attention Mechanism (CAAM). CAAM embeds real-time channel signal-to-noise ratio into the attention computation, enabling hierarchical adaptive weighting of features in different encoder layers. Furthermore, We propose a Frequency Spatial Hybrid Module with two stages, frequency-first followed by spatial processing. Extensive experimental results show that our proposed method significantly enhances image reconstruction quality across two distinct resolution datasets compared to existing methods.