Locally optimum detection schemes for weak composite signals having both deterministic and stochastic signal components in purely-additive noise are derived using the generalized version of the Neyman-Person``s fundamental lemma for three different cases. The locally optimum detector test statistics are derived according to the relative signal strength of the deterministic signal component and the stochastic signal component. A reparametrization rule is proposed to expedite the derivation. The locally optimum detectors for composite signals are compared to those for deterministic signal only and those for random signal only. Schematic diagrams of the locally optimum detector structures are included. Finally examples of the locally optimum detector test statistics are shown for the generalized Gaussian noise distribution, generalized Catchy noise distribution and Student``s t noise distribution.