<p>To meet the growing demand for computing power, photonic computing plays a crucial role in fields such as signal processing and artificial intelligence. However, the noise immunity of existing photonic computing systems still needs improvement due to factors such as noise accumulation and limitations in analog-to-digital converter (ADC) resolution. In this work, we propose a photonic dot-product calculation scheme based on delta-sigma modulation (DSM) to improve computational accuracy. At the transmitter, the proposed scheme converts high-precision data into binary digital signals. Simultaneously, the receiver only requires a low-pass filter to recover the high-precision calculation result. Experimental results show that for dot-product operations on arbitrary 8-bit data, the mean square error (RMSE) of the DSM scheme output is only on the order of 2 × 10⁻². Moreover, this scheme only requires an 8-bit ADC to obtain the complete result. Then, we build a photoelectric hybrid neural network based on the proposed scheme to implement an iris flower classification task. When the system noise level is 5 dB, the classification accuracy of the DSM scheme reaches 93.33%, which is approximately 33% higher than the classical analog scheme. Therefore, the proposed scheme provides an effective approach to realizing a reconfigurable, high-precision photonic computing architecture.</p>