Sarcasm, as a form of implicit emotion, often relies on contextual or multimodal cues for accurate interpretation. However, traditional text-based sarcasm detection models struggle due to the absence of such paralinguistic signals. Observing that emojis are widely used on social media and can significantly enhance or even reverse the emotional polarity of a statement, we consider further mining additional emotional clues contained in emojis to assist in sarcasm detection in text. Specifically, we propose a text-based sarcasm detection model with Emoji Contradictory Clues Assisting (ECCA). To leverage the supplementary information carried by emojis, we explore both the explicit visual images and corresponding textual descriptions of emojis for more comprehensive information. To capture implicit sarcasm more effectively, we design a contrastive attention mechanism at the word level, aligning emoji features with textual tokens to highlight emotional incongruity. Furthermore, a gated fusion module is employed to dynamically control the integration of emoji-text features, and multi-task learning is used to enhance emoji representation learning. Extensive experiments on public datasets prove that emojis are well helpful in sarcastic detection tasks, and fully exploration of emojis can further facilitate capture of contrasting contradictory features implicit in original texts, thus better help detection.