<p>The use of intelligent algorithms for n-<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(\gamma \)</EquationSource> <EquationSource Format="MATHML"><math> <mi>γ</mi> </math></EquationSource> </InlineEquation> discrimination is currently a new research direction. In this paper, the Improved FGA-LDA algorithm is proposed by combining traditional discrimination method with machine learning. Firstly, this paper improves the traditional frequency gradient analysis (FGA) method and defines a brand new discrimination factor. Secondly, this paper combines the Improved FGA method with Linear Discriminant Analysis (LDA) algorithm to realize intelligent n-<InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\gamma \)</EquationSource> <EquationSource Format="MATHML"><math> <mi>γ</mi> </math></EquationSource> </InlineEquation> discrimination. This method utilizes the LDA algorithm to realize feature projection, which can retain the effective information while reducing the feature dimension and improve the classification performance. In order to verify the superiority of the proposed algorithm, two different radioisotope neutron sources(²⁴¹Am-Be and ²⁵²Cf) and a pure gamma source (¹³⁷Cs) are used to validate the proposed algorithm. Throughout the energy domain, the Improved FGA-LDA algorithm has figure of merit (FOM) values of 1.033 and 1.200 for the ²⁴¹Am-Be and ²⁵²Cf neutron source datasets, respectively, demonstrating that it is able to effectively discriminate between neutrons and gamma rays. In the high-energy domain (<InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(E_{p}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>E</mi> <mi>p</mi> </msub> </math></EquationSource> </InlineEquation> <InlineEquation ID="IEq7"> <EquationSource Format="TEX">\(\ge \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>≥</mo> </math></EquationSource> </InlineEquation> 50 keV), the Improved FGA-LDA algorithm improves its FOM value from 1.364 to 1.807 for the ²⁴¹Am-Be dataset, and from 1.534 to 1.820 for the ²⁵²Cf dataset. In the case of using mixed data, the discrimination effect of Improved FGA-LDA algorithm is excellent. The experimental results show that the Improved FGA-LDA algorithm has a better discrimination effect and at the same time can be applied to different mixed neutron and gamma radiation fields.</p>