Abstract: To address the challenges in detection and recognition arising from significant variations in target pixel occupancy within the field of view, an optoelectronic intelligent perception framework based on Dempster-Shafer (DS) evidence theory is proposed in this paper. The implementation of this framework proceeds in two primary stages. First, a multi-dimensional parallel detection mechanism is established to simultaneously execute four heterogeneous algorithms on single-frame imagery. These algorithms encompass prior-knowledge-based and deep-learning-based detection for small targets, large-scale target confirmation, and fine-grained classification, thereby comprehensively capturing the spatiotemporal positions and attribute features of the targets. Subsequently, DS evidence theory is employed as a decision-level fusion strategy to synthesize the aforementioned multi-source heterogeneous evidence, ultimately yielding high-confidence integrated recognition results. Simulation results demonstrate that the target detection and recognition technology achieves a detection probability of 98.5%, a false alarm rate of 1.8%, and a coarse-level classification accuracy of 98%. This technical solution holds significant practical value and offers broad reference significance fir systematically addressing detection and recognition functions in optoeletronic equipment.