The increasing computational demands of machine learning models have driven interest in developing unconventional computing hardware to improve speed and energy efficiency. In this work, we introduce an integrated photonic chip designed to perform convolution operations in deep convolutional neural networks. The convolutional kernel is implemented in the optical circuit, which functions as a two-port lattice filter, by modifying the optical signal paths through phase shifters. Using a simulated model of the optical chip that implements the convolutional layers, we evaluate the performance of a deep convolutional network trained on the CIFAR-10 dataset. We also examine the impact of hardware limitations, such as system noise and quantization, on the model performance.