This study presents a reconfigurable perovskite X-ray detector for intelligent imaging, which integrates computing within the X-ray detector to enhance data processing efficiency and image quality. The detector is based on a CsPbBr₃ single-crystal structure, offering convenient polarity reconfigurability, a high linear dynamic range of 106 dB, and robust stability. The device achieves edge extraction imaging with a data compression ratio of ~50% and can be programmed for pattern recognition tasks with 100% accuracy. The detector's high linearity and stability make it suitable for complex and flexible scenarios, promising applications in intelligent X-ray imaging. X-ray detection is crucial in various fields, including medical imaging, security screening, and industrial inspection. However, advancements in X-ray detectors have led to increased data volumes, posing challenges in data transmission, processing, and storage. Traditional X-ray systems suffer from high power consumption, complexity, and bulkiness. In-sensor computing, inspired by neural network vision sensors, offers a solution by enabling low power consumption and minimal latency in visible light detection. This technology is now being explored for X-ray detection to overcome transmission bandwidth limitations. The proposed detector utilizes a CsPbBr₃ single-crystal with a N-I-P structure, which allows for polarity reconfigurability and linear responsivity. Surface defects were passivated using a C₆₀ layer, significantly improving the detector's performance. The detector exhibits a high linear dynamic range and robust stability, enabling efficient data compression and intelligent image processing. The device can perform edge extraction, data compression, and pattern recognition tasks, demonstrating its potential for intelligent X-ray imaging. The study also highlights the detector's ability to process images using convolutional kernels, achieving a data compression ratio of 46.6% with the Laplacian kernel. The detector's performance was validated through various tests, including edge extraction, data compression, and pattern recognition. The results show that the detector can achieve high accuracy in pattern recognition tasks, making it a promising platform for intelligent X-ray imaging. The integration of the detector with a CMOS die further enhances its performance and applicability in real-world scenarios. The research provides a foundation for future developments in advanced neural network-based X-ray detectors.This study presents a reconfigurable perovskite X-ray detector for intelligent imaging, which integrates computing within the X-ray detector to enhance data processing efficiency and image quality. The detector is based on a CsPbBr₃ single-crystal structure, offering convenient polarity reconfigurability, a high linear dynamic range of 106 dB, and robust stability. The device achieves edge extraction imaging with a data compression ratio of ~50% and can be programmed for pattern recognition tasks with 100% accuracy. The detector's high linearity and stability make it suitable for complex and flexible scenarios, promising applications in intelligent X-ray imaging. X-ray detection is crucial in various fields, including medical imaging, security screening, and industrial inspection. However, advancements in X-ray detectors have led to increased data volumes, posing challenges in data transmission, processing, and storage. Traditional X-ray systems suffer from high power consumption, complexity, and bulkiness. In-sensor computing, inspired by neural network vision sensors, offers a solution by enabling low power consumption and minimal latency in visible light detection. This technology is now being explored for X-ray detection to overcome transmission bandwidth limitations. The proposed detector utilizes a CsPbBr₃ single-crystal with a N-I-P structure, which allows for polarity reconfigurability and linear responsivity. Surface defects were passivated using a C₆₀ layer, significantly improving the detector's performance. The detector exhibits a high linear dynamic range and robust stability, enabling efficient data compression and intelligent image processing. The device can perform edge extraction, data compression, and pattern recognition tasks, demonstrating its potential for intelligent X-ray imaging. The study also highlights the detector's ability to process images using convolutional kernels, achieving a data compression ratio of 46.6% with the Laplacian kernel. The detector's performance was validated through various tests, including edge extraction, data compression, and pattern recognition. The results show that the detector can achieve high accuracy in pattern recognition tasks, making it a promising platform for intelligent X-ray imaging. The integration of the detector with a CMOS die further enhances its performance and applicability in real-world scenarios. The research provides a foundation for future developments in advanced neural network-based X-ray detectors.