Against the backdrop of global energy transition and sustainable development, electric vehicles (EVs) have emerged as a critical direction in the transportation sector due to their environmental advantages. Lithium-ion batteries (LIB), as their core component, have become the mainstream choice for EVs. However, the associated safety issues of thermal runaway are drawing increasing attention. Currently, data-driven methods based on signal processing have achieved significant progress in the field of LIB fault diagnosis. However, existing studies face challenges, as 1-dimensional (1-D) signals struggle to capture sensitive fluctuations in the early stages of thermal runaway, while 2-D images suffer from information redundancy. To address these limitations, this paper proposes a thermal runaway early warning method for vehicle LIB based on 3-D feature landscape, aiming to ensure the safe and stable operation of EVs and power batteries. This method overcomes the constraints of traditional low-dimensional feature spaces. First, the generalized s-transform is employed to convert 1-D temperature signals into 2-D spectrograms. On this basis, scale invariant features are extracted to construct 3-D feature landscape and generate high information density state barcode. Finally, various machine learning methods are applied to classify the barcode data, enabling accurate identification of thermal runaway states. The proposed algorithm is evaluated using standardized datasets from Oak Ridge National Laboratory, validating its effectiveness. The results demonstrate that the signal processing method based on 3-D feature landscape significantly improves classification accuracy, with the support vector machine model achieving a precision rate of 98.10%.