An improved Lorenz-like system was constructed for weak harmonic signal recognition in strong noise. Applying the law of large numbers, the thresholds were calculated from the bifurcation diagrams, and the mode of statistical data was taken as the critical value. The influence of the noise on the thresholds was also revealed. The amplitude of the weak harmonic signal was identified by the threshold difference of the change in the number of attractors detected before and after adding the weak harmonic signal to the system. The results showed that the reliability and accuracy was 89.5% and the minimum signal-to-noise ratio(SNR) was -33dB. Compared to the Duffing system and the controlled Lorenz-like system, the improved Lorenz-like system had the advantages of higher threshold reliability, lower SNR and lower detection threshold, which was more suitable for identifying signal parameters in strong noise. Additionally, the theoretical results were validated by the analog circuit of the improved Lorenz-like system created by Multisim software.