In detecting methane concentration by tunable laser absorption spectroscopy (TDLAS), the second harmonic signal amplitude of methane transmitted light intensity is directly proportional to the concentration of trace methane gas. How to accurately and quickly screen the amplitude of the second harmonic signal of the target methane transmitted light intensity is crucial. The photodetector obtains the 1 000 methane gas transmitted light intensity signal samples, and it is demodulated to obtain the second harmonic signal. When obtaining a variety of trace methane gas transmitted light intensity and demodulating the second harmonic signal by transmitted light intensity, noise and artificial operation affect the amplitude of the second harmonic signal，resulting in an increase in the time for manual screening of the second harmonic signal. Using traditional TDLAS technology to screen trace methane second harmonic signals had the problem of high time cost. A trace methane concentration detection method based on wide convolution and wide kernel 1D convolutional neural networks (1D-WCWKCNN) was proposed. Firstly, the 1D-WCWKCNN model is trained with the help of the methane gas dataset, and the model parameters are continuously adjusted according to the training results. Secondly, the method used a wide convolution layer and wide convolution kernel 1D convolution layer to extract the features of the trace methane second harmonic signal so that the network obtained the characteristic relationship between a longer sequence and the sequence boundary information in the methane concentration signal and the gas concentration after one convolution. The second harmonic signal of methane transmitted light intensity is extracted through the 6-layer convolutional layer to extract the main characteristics of the relationship between the signal and methane gas concentration.The 6-layer maximal pooling layer retains the main characteristics. The Flatten layer processes the signal data processed by the previous layer in one dimension. Finally, the trained 1D-WCWKCNN model outputs trace methane gas concentration through the Dense layer. The trace methane gas concentration detection model based on 1D-WCWKCNN replaces manually screening second harmonic signals for detecting trace methane gas concentration in a fitted straight line in TDLAS technology. The effectiveness of this method is verified in actual experiments. The results show that it can effectively detect the concentration of trace methane in 50~1 000 mg·L-1, and its accuracy reaches 99.85%. Compared with other methods, it has strong signal feature extraction ability and high detection accuracy of methane gas.This method facilitates the screening of gas concentration signals to be measured in the field of gas detection.