Traditional methods for detecting weak targets predominantly depend on the contrast between the target and its surroundings within a single frame or utilize motion models with static parameters for correlating information across multiple frames. In scenarios where the signal-to-clutter ratio (SCR) is low, these strategies frequently falter, resulting in reduced resilience, suboptimal detection rates, and a heightened likelihood of false positives. To mitigate these issues, the present study puts forth an innovative detection modality for weak targets, fundamentally based on Long Short-Term Memory (LSTM) networks. By incorporating a Convolutional LSTM (ConvLSTM) framework, this modality adeptly assimilates the movement patterns of weak targets across image sequences, which may exhibit sporadic visibility. Extensive simulation experiments affirm that this proposed methodology considerably amplifies both detection and tracking efficacy of weak targets under challenging low SCR conditions, thereby markedly bolstering the precision of detection results. The enhanced detection protocol stands to significantly benefit applications where discerning weak targets from noisy backgrounds is critical, promising substantial advancements in fields such as remote sensing, surveillance, and automated monitoring systems.