Due to the complex underground environment of coal mines, water influx drives the movement of silt and sand, which can cause irregular fluctuations in the water level of the water tank, interfere with water level measurement, and increase the error of detection results. To this end, a mine water level detection method based on multi-sensor fusion and DeepLabv3+algorithm is proposed. Firstly, select ultrasonic liquid level sensors, input type liquid level sensors, temperature sensors, and CCD visual sensors to obtain data on the changes in the water level height of the mine water tank; Then, frequency domain feature correction is applied to the sensing signal, and the mixed Kalman particle filter algorithm is used to fuse the collected multi-sensor data; Finally, using the DeepLabv3+algorithm to process the fused data, a water level detection model based on DeepLabv3+is constructed to achieve water level detection. Through experiments, it is known that whether in the stage of rising or falling water level, the proposed method's detection results for water level height are very close to the actual results, with an error of no more than 0.2 m. The detection range can reach over 9.7 m3, and the detection accuracy can be maintained at 78~89 %. Under the condition of gradually increasing interference values, the error for water level detection in the water tank is 0.21~0.35 m, with high detection accuracy and good application effect.