Résumé : This work focuses on the gas chemical composition of CGCG007-025. This compact dwarf is undergoing a galaxy wide star forming burst, whose spatial behaviour has been observed by VLT/MUSE. We present a new line measurement library to treat almost 7800 voxels. The direct method chemical analysis is limited to 484 voxels with good detection of the $[SIII]$6312$\mathring{\mathrm{A}}$ temperature diagnostic line. The recombination fluxes are corrected for stellar absorption via a population synthesis. Additionally, we discuss a new algorithm to fit photoionization models via neural networks. The 8 ionic abundances analyzed show a spatial normal distribution with a $\sigma\sim0.1\,dex$, where only half this value can be explained by the uncertainty in the measurements. The oxygen abundance distribution is $12+log(O/H)=7.88\pm0.11$. The $T_{e}[SIII]$ and $ne[SII]$ are also normally distributed. However, in the central and brightest region, the $ne[SII]$ is almost thrice the mean galaxy value. This is also reflected in the extinction measurements. The ionization parameter has a distribution of $log(U) = -2.52^{0.17}_{0.19}$. The parameter spatial behaviour agrees with the $S^{2+}/S^{+}$ map. Finally, the discrepancies between the direct method and the photoionization model fitting are discussed. In the latter technique, we find that mixing lines with uneven uncertainty magnitudes can impact the accuracy of the results. In these fittings, we recommend overestimating the minimum flux uncertainty one order below the maximum line flux uncertainty. This provides a better match with the direct method.