Abstract: With the advancement of large-scale and standardized aquaculture, Long Range Radio (LoRa) technology, leveraging its advantages of low cost, low power consumption, long transmission distance, and independence from network operators, provides a low-power, low-cost communication solution for the Fishery Internet of Things (IoT), leading to its wider application. Furthermore, the complex environment of large outdoor farming areas and seasonal variations in air temperature and humidity affect the stability of wireless signals to varying degrees. Moreover, the scaling of aquaculture leads to a sharp increase in the number of IoT devices, making data collision problems more prominent and posing multiple challenges to wireless network performance. Addressing these issues, this research fully leverages the multi-Spreading Factor (SF) characteristic of LoRa to design a LoRa multi-channel communication mechanism tailored for fishery IoT. Building on this foundation, a joint Spreading Factor and time slot allocation strategy, termed AOA-CB-SS, is proposed. This strategy interleaves SF allocation for different node types to avoid mutual interference between the communication links of sensor nodes and controller nodes, thereby enhancing the stability and reliability of the network system. First, node type, Received Signal Strength Indicator (RSSI), and Signal-to-Noise Ratio (SNR) are used as inputs to a CatBoost (CB) model. The Arithmetic Optimization Algorithm (AOA) is employed to optimize the hyperparameters of this model, enabling more accurate and rapid allocation of Spreading Factors (SF) to the nodes. Subsequently, based on the current SF distribution within the network, a hierarchical Slot Selection (SS) strategy is utilized to assign time slots to sensor nodes, ensuring balanced network resource allocation. Through the SS strategy, the network allows multiple devices with different Spreading Factors to communicate simultaneously within the same time frame, while devices with the same Spreading Factor communicate in different time slots. Simulation and experimental results demonstrate that the AOA-CB-SS strategy achieves an SF allocation accuracy of 98%. Compared to similar strategies, its Packet Delivery Ratio (PDR) is increased by over 9.89%, and average energy consumption is reduced by 16.8%. This strategy permits nodes using different SFs to reuse time slots, thereby reducing packet collisions while simultaneously increasing wireless bandwidth utilization and enhancing the scalability of the network system. Field tests showed an average PDR of 96.7% for sensor nodes. For controller nodes, the success rate for single-attempt transmissions of gateway control messages reached 96.5%. This research offers new perspectives for the application of large-scale fishery IoT. Furthermore, this approach can be effectively extended to other agricultural domains (such as large-scale fruit and vegetable cultivation) employing LoRa IoT systems. This requires only training the model with datasets specific to the application scenario and modifying the upload period, time slot length, and number in the SS component according to the application's requirements.