Moisture content critically affects the stability, nutritional value, and shelf life of pellet feed during production and storage, yet conventional measurement methods are limited by slow response, poor accuracy, strong temperature sensitivity, and poor on-site applicability. To overcome these issues, this study proposes a dielectric-based detection method incorporating electrode geometry optimization and temperature-adaptive compensation, using sheep pellet feed as the test material. COMSOL Multiphysics simulations revealed the effects of electrode thickness, spacing, and overlap area on the fringe field; the optimal configuration of 1 mm thickness, 40 mm spacing, and 15,000 mm² overlap area reduced fringe field width by 32.7% and improved field uniformity by 41.2%. A cubic polynomial coupling model relating temperature, moisture content, and relative permittivity was established via a two-factor, five-level full factorial design with polynomial response surface regression, yielding a coefficient of determination of 0.996, an adjusted R² of 0.994, and a root-mean-square error of 0.123, thereby enabling temperature-adaptive compensation. An STM32F103VET6-based prototype with capacitance sensing, temperature acquisition, display, and storage was developed. Across moisture contents of 10% to 20% and temperatures of 15 to 35 °C, it achieves a mean absolute error of 0.350%, a mean relative error of 2.600%, a repeatability coefficient of variation of 1.120%, and a one-hour drift below 0.080%. The device offers a fast, low-cost, and interference-resistant solution for real-time feed quality monitoring.