The long-term monitoring of respiratory status is crucial for the prevention and diagnosis of respiratory diseases. However, existing continuous respiratory monitoring devices are typically bulky and require either chest strapping or proximity to the nasal area, which compromises user comfort and may disrupt the monitoring process. To overcome these challenges, we have developed a flexible, attachable, lightweight, and miniaturized system designed for extended wear on the wrist. This system incorporates signal acquisition circuitry, a mobile client, and a deep neural network, facilitating long-term respiratory monitoring. Specifically, we fabricated a highly sensitive (11,847.24 kPa) flexible pressure sensor using a screen printing process, which is capable of functioning beyond 70,000 cycles. Additionally, we engineered a bidirectional long short-term memory (BiLSTM) neural network, enhanced with a residual module, to classify various respiratory states including slow, normal, fast, and simulated breathing. The system achieved a dataset classification accuracy exceeding 99.5%. We have successfully demonstrated a stable, cost-effective, and durable respiratory sensor system that can quantitatively collect and store respiratory data for individuals and groups. This system holds potential for everyday monitoring of physiological signals and healthcare applications.