Abstract: In view of the uncertain mechanical properties of knot-bearing wood and the difficulty of judging whether it is usable, this article proposes a method to evaluate the usability of wood containing knots by detecting the bending properties of wood containing knots. The common tree Quercus mongolica, which accounts for 15%-20% of the total forest area in Northeast China, is selected as the experimental object. Firstly, the object detection algorithm is used to identify the areas containing knots on the suface of wood, followed by spectral extraction of the identified area and the construction of a quantitative prediction model. Finally, the mechanical properties of wood containing knots are analyzed through deep learning. The experimental results indicate that the SPA-SVM prediction model proposed in this article has excellent predictive ability for the bending properties of wood, with experimental results indicators of R2=0. 96, RMSE=0. 58, and RPD=5. 09. The prediction model proposed in this article can accurately predict the bending properties of wood containing knots. The predicted results have a small error with the actual values, which meets the experimental requirements and standards. The predicted results can provide a basis for whether the wood can be used.