Machine vision, serving as the “eyes” of artificial intelligence (AI), is one of the key windows for AI to acquire external information. However, traditional machine vision relies on the Von Neumann architecture, where sensing, storage, and processing are separated. This architecture necessitates constant data transfer between different units, inevitably leading to high power consumption and latency. To address these challenges, a PtSe₂ photosynaptic device with negative light response is prepared. The device shows an inhibitory postsynaptic current (IPSC) under light pulse stimulation, and achieves optically tunable synaptic behaviors, including double pulse facilitation (PPD), short-range plasticity (STP), and long-range plasticity (LTP). In addition, by using a 3 × 3 sensor array, the device exhibits dependence on light duration, and the image in-situ sensing and storage functions are demonstrated and verified. By using 28 × 28 device array combined with artificial neural network (ANN), the integrated perception-storage-preprocessing function of visual information is realized. The experimental results show that the image after preprocessing (denoising) is trained for 100 epochs, and the accuracy rate reaches 91%. Finally, lasers with two representative wavelengths of 405 nm and 532 nm are chosen as the light sources in the experiment, and the I-V characteristic curve changes most under the blue light pulse of 450 nm, which is because the blue light has higher photon energy to produce negative light effect. Based on the different photocurrents of the device responding to different wavelengths of light, the photoelectric synaptic logic gates ‘NOR’, ‘NAND’ and ‘XOR’ are established, which enables image processing functions such as dilation, erosion and difference recognition. The device’s power consumption is calculated to be 0.111 nJ per spike. The research results indicate that the negative photoconductivity of PtSe₂ has great potential in simplifying information processing and effectively promoting applications, which will help promote more integrated and efficient NVS.