Some boat noises are perceived as noisy and annoying by people who live near canals. We have previously proposed an information system that uses audiovisual means to provide cellphone alerts of approaching noisy boats, but a problem with that system was camera-based detection of a boat approaching at night. In the present paper, we investigate using training data to detect boat noise in environmental sound by means of a convolutional neural network. To detect boat noise, training data are used involving spectrograms of the environmental sound. The spectrogram configuration is investigated to improve the detection of boat noise. From the results, when the spectrogram configuration has a time axis of 5 s and a frequency axis of 10-3,500 Hz, the detection performance has a highest accuracy of over 95%.